"Merge Tagged PR 4836"

common: fs: More misc. changes

src/CMakeLists.txt

@@ -45,14 +45,23 @@ if (MSVC)
         /Zc:inline
         /Zc:throwingNew
 
+        # External headers diagnostics
+        /experimental:external  # Enables the external headers options. This option isn't required in Visual Studio 2019 version 16.10 and later
+        /external:anglebrackets # Treats all headers included by #include <header>, where the header file is enclosed in angle brackets (< >), as external headers
+        /external:W0            # Sets the default warning level to 0 for external headers, effectively turning off warnings for external headers
+
         # Warnings
         /W3
-        /we4062 # enumerator 'identifier' in a switch of enum 'enumeration' is not handled
+        /we4018 # 'expression': signed/unsigned mismatch
+        /we4062 # Enumerator 'identifier' in a switch of enum 'enumeration' is not handled
         /we4101 # 'identifier': unreferenced local variable
         /we4189 # 'identifier': local variable is initialized but not referenced
         /we4265 # 'class': class has virtual functions, but destructor is not virtual
-        /we4388 # signed/unsigned mismatch
-        /we4547 # 'operator' : operator before comma has no effect; expected operator with side-effect
+        /we4267 # 'var': conversion from 'size_t' to 'type', possible loss of data
+        /we4305 # 'context': truncation from 'type1' to 'type2'
+        /we4388 # 'expression': signed/unsigned mismatch
+        /we4389 # 'operator': signed/unsigned mismatch
+        /we4547 # 'operator': operator before comma has no effect; expected operator with side-effect
         /we4549 # 'operator1': operator before comma has no effect; did you intend 'operator2'?
         /we4555 # Expression has no effect; expected expression with side-effect
         /we4715 # 'function': not all control paths return a value
@@ -73,6 +82,7 @@ else()
         -Werror=missing-declarations
         -Werror=missing-field-initializers
         -Werror=reorder
+        -Werror=sign-compare
         -Werror=switch
         -Werror=uninitialized
         -Werror=unused-function

src/audio_core/CMakeLists.txt

@@ -51,9 +51,6 @@ if (NOT MSVC)
     target_compile_options(audio_core PRIVATE
         -Werror=conversion
         -Werror=ignored-qualifiers
-        -Werror=implicit-fallthrough
-        -Werror=reorder
-        -Werror=sign-compare
         -Werror=shadow
         -Werror=unused-parameter
         -Werror=unused-variable

src/common/fs/file.cpp

@@ -306,9 +306,9 @@ bool IOFile::Flush() const {
     errno = 0;
 
 #ifdef _WIN32
-    const auto flush_result = std::fflush(file) == 0 && _commit(fileno(file)) == 0;
+    const auto flush_result = std::fflush(file) == 0;
 #else
-    const auto flush_result = std::fflush(file) == 0 && fsync(fileno(file)) == 0;
+    const auto flush_result = std::fflush(file) == 0;
 #endif
 
     if (!flush_result) {
@@ -320,6 +320,28 @@ bool IOFile::Flush() const {
     return flush_result;
 }
 
+bool IOFile::Commit() const {
+    if (!IsOpen()) {
+        return false;
+    }
+
+    errno = 0;
+
+#ifdef _WIN32
+    const auto commit_result = std::fflush(file) == 0 && _commit(fileno(file)) == 0;
+#else
+    const auto commit_result = std::fflush(file) == 0 && fsync(fileno(file)) == 0;
+#endif
+
+    if (!commit_result) {
+        const auto ec = std::error_code{errno, std::generic_category()};
+        LOG_ERROR(Common_Filesystem, "Failed to commit the file at path={}, ec_message={}",
+                  PathToUTF8String(file_path), ec.message());
+    }
+
+    return commit_result;
+}
+
 bool IOFile::SetSize(u64 size) const {
     if (!IsOpen()) {
         return false;
@@ -347,6 +369,9 @@ u64 IOFile::GetSize() const {
         return 0;
     }
 
+    // Flush any unwritten buffered data into the file prior to retrieving the file size.
+    std::fflush(file);
+
     std::error_code ec;
 
     const auto file_size = fs::file_size(file_path, ec);

src/common/fs/file.h

@@ -396,12 +396,21 @@ public:
     [[nodiscard]] size_t WriteString(std::span<const char> string) const;
 
     /**
-     * Attempts to flush any unwritten buffered data into the file and flush the file into the disk.
+     * Attempts to flush any unwritten buffered data into the file.
      *
      * @returns True if the flush was successful, false otherwise.
      */
     bool Flush() const;
 
+    /**
+     * Attempts to commit the file into the disk.
+     * Note that this is an expensive operation as this forces the operating system to write
+     * the contents of the file associated with the file descriptor into the disk.
+     *
+     * @returns True if the commit was successful, false otherwise.
+     */
+    bool Commit() const;
+
     /**
      * Resizes the file to a given size.
      * If the file is resized to a smaller size, the remainder of the file is discarded.

src/common/logging/backend.cpp

@@ -171,19 +171,22 @@ FileBackend::FileBackend(const std::filesystem::path& filename) {
 FileBackend::~FileBackend() = default;
 
 void FileBackend::Write(const Entry& entry) {
-    using namespace Common::Literals;
-    // prevent logs from going over the maximum size (in case its spamming and the user doesn't
-    // know)
-    constexpr std::size_t MAX_BYTES_WRITTEN = 100_MiB;
-    constexpr std::size_t MAX_BYTES_WRITTEN_EXTENDED = 1_GiB;
-
     if (!file->IsOpen()) {
         return;
     }
 
-    if (Settings::values.extended_logging && bytes_written > MAX_BYTES_WRITTEN_EXTENDED) {
-        return;
-    } else if (!Settings::values.extended_logging && bytes_written > MAX_BYTES_WRITTEN) {
+    using namespace Common::Literals;
+    // Prevent logs from exceeding a set maximum size in the event that log entries are spammed.
+    constexpr std::size_t MAX_BYTES_WRITTEN = 100_MiB;
+    constexpr std::size_t MAX_BYTES_WRITTEN_EXTENDED = 1_GiB;
+
+    const bool write_limit_exceeded =
+        bytes_written > MAX_BYTES_WRITTEN_EXTENDED ||
+        (bytes_written > MAX_BYTES_WRITTEN && !Settings::values.extended_logging);
+
+    // Close the file after the write limit is exceeded.
+    if (write_limit_exceeded) {
+        file->Close();
         return;
     }
 

src/core/CMakeLists.txt

@@ -654,24 +654,19 @@ endif()
 
 if (MSVC)
     target_compile_options(core PRIVATE
-        /we4018 # 'expression' : signed/unsigned mismatch
-        /we4244 # 'argument' : conversion from 'type1' to 'type2', possible loss of data (floating-point)
-        /we4245 # 'conversion' : conversion from 'type1' to 'type2', signed/unsigned mismatch
+        /we4242 # 'identifier': conversion from 'type1' to 'type2', possible loss of data
+        /we4244 # 'conversion': conversion from 'type1' to 'type2', possible loss of data
+        /we4245 # 'conversion': conversion from 'type1' to 'type2', signed/unsigned mismatch
         /we4254 # 'operator': conversion from 'type1:field_bits' to 'type2:field_bits', possible loss of data
-        /we4267 # 'var' : conversion from 'size_t' to 'type', possible loss of data
-        /we4305 # 'context' : truncation from 'type1' to 'type2'
         /we4456 # Declaration of 'identifier' hides previous local declaration
         /we4457 # Declaration of 'identifier' hides function parameter
         /we4458 # Declaration of 'identifier' hides class member
         /we4459 # Declaration of 'identifier' hides global declaration
-        /we4715 # 'function' : not all control paths return a value
     )
 else()
     target_compile_options(core PRIVATE
         -Werror=conversion
         -Werror=ignored-qualifiers
-        -Werror=implicit-fallthrough
-        -Werror=sign-compare
         -Werror=shadow
 
         $<$<CXX_COMPILER_ID:GNU>:-Werror=class-memaccess>

src/core/hle/service/mii/manager.cpp

@@ -20,6 +20,7 @@ namespace {
 
 constexpr ResultCode ERROR_CANNOT_FIND_ENTRY{ErrorModule::Mii, 4};
 
+constexpr std::size_t BaseMiiCount{2};
 constexpr std::size_t DefaultMiiCount{RawData::DefaultMii.size()};
 
 constexpr MiiStoreData::Name DefaultMiiName{u'y', u'u', u'z', u'u'};
@@ -415,7 +416,7 @@ u32 MiiManager::GetCount(SourceFlag source_flag) const {
         count += 0;
     }
     if ((source_flag & SourceFlag::Default) != SourceFlag::None) {
-        count += DefaultMiiCount;
+        count += (DefaultMiiCount - BaseMiiCount);
     }
     return static_cast<u32>(count);
 }
@@ -445,7 +446,7 @@ ResultVal<std::vector<MiiInfoElement>> MiiManager::GetDefault(SourceFlag source_
         return MakeResult(std::move(result));
     }
 
-    for (std::size_t index = 0; index < DefaultMiiCount; index++) {
+    for (std::size_t index = BaseMiiCount; index < DefaultMiiCount; index++) {
         result.emplace_back(BuildDefault(index), Source::Default);
     }
 

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

@@ -68,6 +68,7 @@ void nvhost_nvdec::OnOpen(DeviceFD fd) {}
 void nvhost_nvdec::OnClose(DeviceFD fd) {
     LOG_INFO(Service_NVDRV, "NVDEC video stream ended");
     system.GPU().ClearCdmaInstance();
+    system.GPU().MemoryManager().InvalidateQueuedCaches();
 }
 
 } // namespace Service::Nvidia::Devices

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

@@ -198,7 +198,13 @@ NvResult nvhost_nvdec_common::UnmapBuffer(const std::vector<u8>& input, std::vec
             return NvResult::InvalidState;
         }
         if (const auto size{RemoveBufferMap(object->dma_map_addr)}; size) {
-            gpu.MemoryManager().Unmap(object->dma_map_addr, *size);
+            if (vic_device) {
+                // UnmapVicFrame defers texture_cache invalidation of the frame address until
+                // the stream is over
+                gpu.MemoryManager().UnmapVicFrame(object->dma_map_addr, *size);
+            } else {
+                gpu.MemoryManager().Unmap(object->dma_map_addr, *size);
+            }
         } else {
             // This occurs quite frequently, however does not seem to impact functionality
             LOG_DEBUG(Service_NVDRV, "invalid offset=0x{:X} dma=0x{:X}", object->addr,

src/core/hle/service/nvdrv/devices/nvhost_nvdec_common.h

@@ -160,6 +160,7 @@ protected:
 
     s32_le nvmap_fd{};
     u32_le submit_timeout{};
+    bool vic_device{};
     std::shared_ptr<nvmap> nvmap_dev;
     SyncpointManager& syncpoint_manager;
     std::array<u32, MaxSyncPoints> device_syncpoints{};

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

@@ -12,8 +12,9 @@
 namespace Service::Nvidia::Devices {
 nvhost_vic::nvhost_vic(Core::System& system_, std::shared_ptr<nvmap> nvmap_dev_,
                        SyncpointManager& syncpoint_manager_)
-    : nvhost_nvdec_common{system_, std::move(nvmap_dev_), syncpoint_manager_} {}
-
+    : nvhost_nvdec_common(system_, std::move(nvmap_dev_), syncpoint_manager_) {
+    vic_device = true;
+}
 nvhost_vic::~nvhost_vic() = default;
 
 NvResult nvhost_vic::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input,

src/input_common/CMakeLists.txt

@@ -34,28 +34,17 @@ if (MSVC)
         /W4
         /WX
 
-        # 'expression' : signed/unsigned mismatch
-        /we4018
-        # 'argument' : conversion from 'type1' to 'type2', possible loss of data (floating-point)
-        /we4244
-        # 'conversion' : conversion from 'type1' to 'type2', signed/unsigned mismatch
-        /we4245
-        # 'operator': conversion from 'type1:field_bits' to 'type2:field_bits', possible loss of data
-        /we4254
-        # 'var' : conversion from 'size_t' to 'type', possible loss of data
-        /we4267
-        # 'context' : truncation from 'type1' to 'type2'
-        /we4305
+        /we4242 # 'identifier': conversion from 'type1' to 'type2', possible loss of data
+        /we4244 # 'conversion': conversion from 'type1' to 'type2', possible loss of data
+        /we4245 # 'conversion': conversion from 'type1' to 'type2', signed/unsigned mismatch
+        /we4254 # 'operator': conversion from 'type1:field_bits' to 'type2:field_bits', possible loss of data
     )
 else()
     target_compile_options(input_common PRIVATE
         -Werror
         -Werror=conversion
         -Werror=ignored-qualifiers
-        -Werror=implicit-fallthrough
-        -Werror=reorder
         -Werror=shadow
-        -Werror=sign-compare
         $<$<CXX_COMPILER_ID:GNU>:-Werror=unused-but-set-parameter>
         $<$<CXX_COMPILER_ID:GNU>:-Werror=unused-but-set-variable>
         -Werror=unused-variable

src/input_common/analog_from_button.cpp

@@ -27,6 +27,7 @@ public:
         down->SetCallback(callbacks);
         left->SetCallback(callbacks);
         right->SetCallback(callbacks);
+        modifier->SetCallback(callbacks);
     }
 
     bool IsAngleGreater(float old_angle, float new_angle) const {

src/input_common/gcadapter/gc_adapter.cpp

@@ -5,14 +5,7 @@
 #include <chrono>
 #include <thread>
 
-#ifdef _MSC_VER
-#pragma warning(push)
-#pragma warning(disable : 4200) // nonstandard extension used : zero-sized array in struct/union
-#endif
 #include <libusb.h>
-#ifdef _MSC_VER
-#pragma warning(pop)
-#endif
 
 #include "common/logging/log.h"
 #include "common/param_package.h"

src/input_common/udp/protocol.h

@@ -8,14 +8,7 @@
 #include <optional>
 #include <type_traits>
 
-#ifdef _MSC_VER
-#pragma warning(push)
-#pragma warning(disable : 4701)
-#endif
 #include <boost/crc.hpp>
-#ifdef _MSC_VER
-#pragma warning(pop)
-#endif
 
 #include "common/bit_field.h"
 #include "common/swap.h"

src/video_core/CMakeLists.txt

@@ -292,13 +292,12 @@ endif()
 
 if (MSVC)
     target_compile_options(video_core PRIVATE
-        /we4267 # 'var' : conversion from 'size_t' to 'type', possible loss of data
-        /we4244 # 'var' : conversion from integer to 'type', possible loss of data
+        /we4242 # 'identifier': conversion from 'type1' to 'type2', possible loss of data
+        /we4244 # 'conversion': conversion from 'type1' to 'type2', possible loss of data
         /we4456 # Declaration of 'identifier' hides previous local declaration
         /we4457 # Declaration of 'identifier' hides function parameter
         /we4458 # Declaration of 'identifier' hides class member
         /we4459 # Declaration of 'identifier' hides global declaration
-        /we4715 # 'function' : not all control paths return a value
     )
 else()
     target_compile_options(video_core PRIVATE

src/video_core/cdma_pusher.cpp

@@ -103,8 +103,7 @@ void CDmaPusher::ExecuteCommand(u32 state_offset, u32 data) {
         case ThiMethod::SetMethod1:
             LOG_DEBUG(Service_NVDRV, "NVDEC method 0x{:X}",
                       static_cast<u32>(nvdec_thi_state.method_0));
-            nvdec_processor->ProcessMethod(static_cast<Nvdec::Method>(nvdec_thi_state.method_0),
-                                           data);
+            nvdec_processor->ProcessMethod(nvdec_thi_state.method_0, data);
             break;
         default:
             break;

src/video_core/command_classes/codecs/codec.cpp

@@ -23,8 +23,8 @@ void AVFrameDeleter(AVFrame* ptr) {
     av_free(ptr);
 }
 
-Codec::Codec(GPU& gpu_)
-    : gpu(gpu_), h264_decoder(std::make_unique<Decoder::H264>(gpu)),
+Codec::Codec(GPU& gpu_, const NvdecCommon::NvdecRegisters& regs)
+    : gpu(gpu_), state{regs}, h264_decoder(std::make_unique<Decoder::H264>(gpu)),
       vp9_decoder(std::make_unique<Decoder::VP9>(gpu)) {}
 
 Codec::~Codec() {
@@ -43,46 +43,48 @@ Codec::~Codec() {
     avcodec_close(av_codec_ctx);
 }
 
-void Codec::SetTargetCodec(NvdecCommon::VideoCodec codec) {
-    if (current_codec != codec) {
-        LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", static_cast<u32>(codec));
-        current_codec = codec;
+void Codec::Initialize() {
+    AVCodecID codec{AV_CODEC_ID_NONE};
+    switch (current_codec) {
+    case NvdecCommon::VideoCodec::H264:
+        codec = AV_CODEC_ID_H264;
+        break;
+    case NvdecCommon::VideoCodec::Vp9:
+        codec = AV_CODEC_ID_VP9;
+        break;
+    default:
+        return;
     }
+    av_codec = avcodec_find_decoder(codec);
+    av_codec_ctx = avcodec_alloc_context3(av_codec);
+    av_opt_set(av_codec_ctx->priv_data, "tune", "zerolatency", 0);
+
+    // TODO(ameerj): libavcodec gpu hw acceleration
+
+    const auto av_error = avcodec_open2(av_codec_ctx, av_codec, nullptr);
+    if (av_error < 0) {
+        LOG_ERROR(Service_NVDRV, "avcodec_open2() Failed.");
+        avcodec_close(av_codec_ctx);
+        return;
+    }
+    initialized = true;
+    return;
 }
 
-void Codec::StateWrite(u32 offset, u64 arguments) {
-    u8* const state_offset = reinterpret_cast<u8*>(&state) + offset * sizeof(u64);
-    std::memcpy(state_offset, &arguments, sizeof(u64));
+void Codec::SetTargetCodec(NvdecCommon::VideoCodec codec) {
+    if (current_codec != codec) {
+        current_codec = codec;
+        LOG_INFO(Service_NVDRV, "NVDEC video codec initialized to {}", GetCurrentCodecName());
+    }
 }
 
 void Codec::Decode() {
-    bool is_first_frame = false;
+    const bool is_first_frame = !initialized;
     if (!initialized) {
-        if (current_codec == NvdecCommon::VideoCodec::H264) {
-            av_codec = avcodec_find_decoder(AV_CODEC_ID_H264);
-        } else if (current_codec == NvdecCommon::VideoCodec::Vp9) {
-            av_codec = avcodec_find_decoder(AV_CODEC_ID_VP9);
-        } else {
-            LOG_ERROR(Service_NVDRV, "Unknown video codec {}", current_codec);
-            return;
-        }
-
-        av_codec_ctx = avcodec_alloc_context3(av_codec);
-        av_opt_set(av_codec_ctx->priv_data, "tune", "zerolatency", 0);
-
-        // TODO(ameerj): libavcodec gpu hw acceleration
-
-        const auto av_error = avcodec_open2(av_codec_ctx, av_codec, nullptr);
-        if (av_error < 0) {
-            LOG_ERROR(Service_NVDRV, "avcodec_open2() Failed.");
-            avcodec_close(av_codec_ctx);
-            return;
-        }
-        initialized = true;
-        is_first_frame = true;
+        Initialize();
     }
-    bool vp9_hidden_frame = false;
 
+    bool vp9_hidden_frame = false;
     AVPacket packet{};
     av_init_packet(&packet);
     std::vector<u8> frame_data;
@@ -95,7 +97,7 @@ void Codec::Decode() {
     }
 
     packet.data = frame_data.data();
-    packet.size = static_cast<int>(frame_data.size());
+    packet.size = static_cast<s32>(frame_data.size());
 
     avcodec_send_packet(av_codec_ctx, &packet);
 
@@ -127,4 +129,21 @@ NvdecCommon::VideoCodec Codec::GetCurrentCodec() const {
     return current_codec;
 }
 
+std::string_view Codec::GetCurrentCodecName() const {
+    switch (current_codec) {
+    case NvdecCommon::VideoCodec::None:
+        return "None";
+    case NvdecCommon::VideoCodec::H264:
+        return "H264";
+    case NvdecCommon::VideoCodec::Vp8:
+        return "VP8";
+    case NvdecCommon::VideoCodec::H265:
+        return "H265";
+    case NvdecCommon::VideoCodec::Vp9:
+        return "VP9";
+    default:
+        return "Unknown";
+    }
+};
+
 } // namespace Tegra

src/video_core/command_classes/codecs/codec.h

@@ -14,18 +14,10 @@ extern "C" {
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wconversion"
 #endif
-#ifdef _MSC_VER
-#pragma warning(push)
-#pragma warning(disable : 4242) // conversion from 'type' to 'type', possible loss of data
-#pragma warning(disable : 4244) // conversion from 'type' to 'type', possible loss of data
-#endif
 #include <libavcodec/avcodec.h>
 #if defined(__GNUC__) || defined(__clang__)
 #pragma GCC diagnostic pop
 #endif
-#ifdef _MSC_VER
-#pragma warning(pop)
-#endif
 }
 
 namespace Tegra {
@@ -42,15 +34,15 @@ class VP9;
 
 class Codec {
 public:
-    explicit Codec(GPU& gpu);
+    explicit Codec(GPU& gpu, const NvdecCommon::NvdecRegisters& regs);
     ~Codec();
 
+    /// Initialize the codec, returning success or failure
+    void Initialize();
+
     /// Sets NVDEC video stream codec
     void SetTargetCodec(NvdecCommon::VideoCodec codec);
 
-    /// Populate NvdecRegisters state with argument value at the provided offset
-    void StateWrite(u32 offset, u64 arguments);
-
     /// Call decoders to construct headers, decode AVFrame with ffmpeg
     void Decode();
 
@@ -59,6 +51,8 @@ public:
 
     /// Returns the value of current_codec
     [[nodiscard]] NvdecCommon::VideoCodec GetCurrentCodec() const;
+    /// Return name of the current codec
+    [[nodiscard]] std::string_view GetCurrentCodecName() const;
 
 private:
     bool initialized{};
@@ -68,10 +62,10 @@ private:
     AVCodecContext* av_codec_ctx{nullptr};
 
     GPU& gpu;
+    const NvdecCommon::NvdecRegisters& state;
     std::unique_ptr<Decoder::H264> h264_decoder;
     std::unique_ptr<Decoder::VP9> vp9_decoder;
 
-    NvdecCommon::NvdecRegisters state{};
     std::queue<AVFramePtr> av_frames{};
 };
 

src/video_core/command_classes/codecs/h264.cpp

@@ -45,135 +45,130 @@ H264::~H264() = default;
 
 const std::vector<u8>& H264::ComposeFrameHeader(const NvdecCommon::NvdecRegisters& state,
                                                 bool is_first_frame) {
-    H264DecoderContext context{};
+    H264DecoderContext context;
     gpu.MemoryManager().ReadBlock(state.picture_info_offset, &context, sizeof(H264DecoderContext));
 
-    const s32 frame_number = static_cast<s32>((context.h264_parameter_set.flags >> 46) & 0x1ffff);
+    const s64 frame_number = context.h264_parameter_set.frame_number.Value();
     if (!is_first_frame && frame_number != 0) {
-        frame.resize(context.frame_data_size);
-
+        frame.resize(context.stream_len);
         gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset, frame.data(), frame.size());
-    } else {
-        /// Encode header
-        H264BitWriter writer{};
-        writer.WriteU(1, 24);
-        writer.WriteU(0, 1);
-        writer.WriteU(3, 2);
-        writer.WriteU(7, 5);
-        writer.WriteU(100, 8);
-        writer.WriteU(0, 8);
-        writer.WriteU(31, 8);
-        writer.WriteUe(0);
-        const auto chroma_format_idc =
-            static_cast<u32>((context.h264_parameter_set.flags >> 12) & 3);
-        writer.WriteUe(chroma_format_idc);
-        if (chroma_format_idc == 3) {
-            writer.WriteBit(false);
-        }
-
-        writer.WriteUe(0);
-        writer.WriteUe(0);
-        writer.WriteBit(false); // QpprimeYZeroTransformBypassFlag
-        writer.WriteBit(false); // Scaling matrix present flag
-
-        const auto order_cnt_type = static_cast<u32>((context.h264_parameter_set.flags >> 14) & 3);
-        writer.WriteUe(static_cast<u32>((context.h264_parameter_set.flags >> 8) & 0xf));
-        writer.WriteUe(order_cnt_type);
-        if (order_cnt_type == 0) {
-            writer.WriteUe(context.h264_parameter_set.log2_max_pic_order_cnt);
-        } else if (order_cnt_type == 1) {
-            writer.WriteBit(context.h264_parameter_set.delta_pic_order_always_zero_flag != 0);
-
-            writer.WriteSe(0);
-            writer.WriteSe(0);
-            writer.WriteUe(0);
-        }
-
-        const s32 pic_height = context.h264_parameter_set.pic_height_in_map_units /
-                               (context.h264_parameter_set.frame_mbs_only_flag ? 1 : 2);
-
-        writer.WriteUe(16);
-        writer.WriteBit(false);
-        writer.WriteUe(context.h264_parameter_set.pic_width_in_mbs - 1);
-        writer.WriteUe(pic_height - 1);
-        writer.WriteBit(context.h264_parameter_set.frame_mbs_only_flag != 0);
-
-        if (!context.h264_parameter_set.frame_mbs_only_flag) {
-            writer.WriteBit(((context.h264_parameter_set.flags >> 0) & 1) != 0);
-        }
-
-        writer.WriteBit(((context.h264_parameter_set.flags >> 1) & 1) != 0);
-        writer.WriteBit(false); // Frame cropping flag
-        writer.WriteBit(false); // VUI parameter present flag
-
-        writer.End();
-
-        // H264 PPS
-        writer.WriteU(1, 24);
-        writer.WriteU(0, 1);
-        writer.WriteU(3, 2);
-        writer.WriteU(8, 5);
-
-        writer.WriteUe(0);
-        writer.WriteUe(0);
-
-        writer.WriteBit(context.h264_parameter_set.entropy_coding_mode_flag != 0);
-        writer.WriteBit(false);
-        writer.WriteUe(0);
-        writer.WriteUe(context.h264_parameter_set.num_refidx_l0_default_active);
-        writer.WriteUe(context.h264_parameter_set.num_refidx_l1_default_active);
-        writer.WriteBit(((context.h264_parameter_set.flags >> 2) & 1) != 0);
-        writer.WriteU(static_cast<s32>((context.h264_parameter_set.flags >> 32) & 0x3), 2);
-        s32 pic_init_qp = static_cast<s32>((context.h264_parameter_set.flags >> 16) & 0x3f);
-        pic_init_qp = (pic_init_qp << 26) >> 26;
-        writer.WriteSe(pic_init_qp);
-        writer.WriteSe(0);
-        s32 chroma_qp_index_offset =
-            static_cast<s32>((context.h264_parameter_set.flags >> 22) & 0x1f);
-        chroma_qp_index_offset = (chroma_qp_index_offset << 27) >> 27;
-
-        writer.WriteSe(chroma_qp_index_offset);
-        writer.WriteBit(context.h264_parameter_set.deblocking_filter_control_flag != 0);
-        writer.WriteBit(((context.h264_parameter_set.flags >> 3) & 1) != 0);
-        writer.WriteBit(context.h264_parameter_set.redundant_pic_count_flag != 0);
-        writer.WriteBit(context.h264_parameter_set.transform_8x8_mode_flag != 0);
-
-        writer.WriteBit(true);
-
-        for (s32 index = 0; index < 6; index++) {
-            writer.WriteBit(true);
-            const auto matrix_x4 =
-                std::vector<u8>(context.scaling_matrix_4.begin(), context.scaling_matrix_4.end());
-            writer.WriteScalingList(matrix_x4, index * 16, 16);
-        }
-
-        if (context.h264_parameter_set.transform_8x8_mode_flag) {
-            for (s32 index = 0; index < 2; index++) {
-                writer.WriteBit(true);
-                const auto matrix_x8 = std::vector<u8>(context.scaling_matrix_8.begin(),
-                                                       context.scaling_matrix_8.end());
-
-                writer.WriteScalingList(matrix_x8, index * 64, 64);
-            }
-        }
-
-        s32 chroma_qp_index_offset2 =
-            static_cast<s32>((context.h264_parameter_set.flags >> 27) & 0x1f);
-        chroma_qp_index_offset2 = (chroma_qp_index_offset2 << 27) >> 27;
-
-        writer.WriteSe(chroma_qp_index_offset2);
-
-        writer.End();
-
-        const auto& encoded_header = writer.GetByteArray();
-        frame.resize(encoded_header.size() + context.frame_data_size);
-        std::memcpy(frame.data(), encoded_header.data(), encoded_header.size());
-
-        gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset,
-                                      frame.data() + encoded_header.size(),
-                                      context.frame_data_size);
+        return frame;
     }
 
+    // Encode header
+    H264BitWriter writer{};
+    writer.WriteU(1, 24);
+    writer.WriteU(0, 1);
+    writer.WriteU(3, 2);
+    writer.WriteU(7, 5);
+    writer.WriteU(100, 8);
+    writer.WriteU(0, 8);
+    writer.WriteU(31, 8);
+    writer.WriteUe(0);
+    const u32 chroma_format_idc =
+        static_cast<u32>(context.h264_parameter_set.chroma_format_idc.Value());
+    writer.WriteUe(chroma_format_idc);
+    if (chroma_format_idc == 3) {
+        writer.WriteBit(false);
+    }
+
+    writer.WriteUe(0);
+    writer.WriteUe(0);
+    writer.WriteBit(false); // QpprimeYZeroTransformBypassFlag
+    writer.WriteBit(false); // Scaling matrix present flag
+
+    writer.WriteUe(static_cast<u32>(context.h264_parameter_set.log2_max_frame_num_minus4.Value()));
+
+    const auto order_cnt_type =
+        static_cast<u32>(context.h264_parameter_set.pic_order_cnt_type.Value());
+    writer.WriteUe(order_cnt_type);
+    if (order_cnt_type == 0) {
+        writer.WriteUe(context.h264_parameter_set.log2_max_pic_order_cnt_lsb_minus4);
+    } else if (order_cnt_type == 1) {
+        writer.WriteBit(context.h264_parameter_set.delta_pic_order_always_zero_flag != 0);
+
+        writer.WriteSe(0);
+        writer.WriteSe(0);
+        writer.WriteUe(0);
+    }
+
+    const s32 pic_height = context.h264_parameter_set.frame_height_in_map_units /
+                           (context.h264_parameter_set.frame_mbs_only_flag ? 1 : 2);
+
+    writer.WriteUe(16);
+    writer.WriteBit(false);
+    writer.WriteUe(context.h264_parameter_set.pic_width_in_mbs - 1);
+    writer.WriteUe(pic_height - 1);
+    writer.WriteBit(context.h264_parameter_set.frame_mbs_only_flag != 0);
+
+    if (!context.h264_parameter_set.frame_mbs_only_flag) {
+        writer.WriteBit(context.h264_parameter_set.flags.mbaff_frame.Value() != 0);
+    }
+
+    writer.WriteBit(context.h264_parameter_set.flags.direct_8x8_inference.Value() != 0);
+    writer.WriteBit(false); // Frame cropping flag
+    writer.WriteBit(false); // VUI parameter present flag
+
+    writer.End();
+
+    // H264 PPS
+    writer.WriteU(1, 24);
+    writer.WriteU(0, 1);
+    writer.WriteU(3, 2);
+    writer.WriteU(8, 5);
+
+    writer.WriteUe(0);
+    writer.WriteUe(0);
+
+    writer.WriteBit(context.h264_parameter_set.entropy_coding_mode_flag != 0);
+    writer.WriteBit(false);
+    writer.WriteUe(0);
+    writer.WriteUe(context.h264_parameter_set.num_refidx_l0_default_active);
+    writer.WriteUe(context.h264_parameter_set.num_refidx_l1_default_active);
+    writer.WriteBit(context.h264_parameter_set.flags.weighted_pred.Value() != 0);
+    writer.WriteU(static_cast<s32>(context.h264_parameter_set.weighted_bipred_idc.Value()), 2);
+    s32 pic_init_qp = static_cast<s32>(context.h264_parameter_set.pic_init_qp_minus26.Value());
+    writer.WriteSe(pic_init_qp);
+    writer.WriteSe(0);
+    s32 chroma_qp_index_offset =
+        static_cast<s32>(context.h264_parameter_set.chroma_qp_index_offset.Value());
+
+    writer.WriteSe(chroma_qp_index_offset);
+    writer.WriteBit(context.h264_parameter_set.deblocking_filter_control_present_flag != 0);
+    writer.WriteBit(context.h264_parameter_set.flags.constrained_intra_pred.Value() != 0);
+    writer.WriteBit(context.h264_parameter_set.redundant_pic_cnt_present_flag != 0);
+    writer.WriteBit(context.h264_parameter_set.transform_8x8_mode_flag != 0);
+
+    writer.WriteBit(true);
+
+    for (s32 index = 0; index < 6; index++) {
+        writer.WriteBit(true);
+        std::span<const u8> matrix{context.weight_scale};
+        writer.WriteScalingList(matrix, index * 16, 16);
+    }
+
+    if (context.h264_parameter_set.transform_8x8_mode_flag) {
+        for (s32 index = 0; index < 2; index++) {
+            writer.WriteBit(true);
+            std::span<const u8> matrix{context.weight_scale_8x8};
+            writer.WriteScalingList(matrix, index * 64, 64);
+        }
+    }
+
+    s32 chroma_qp_index_offset2 =
+        static_cast<s32>(context.h264_parameter_set.second_chroma_qp_index_offset.Value());
+
+    writer.WriteSe(chroma_qp_index_offset2);
+
+    writer.End();
+
+    const auto& encoded_header = writer.GetByteArray();
+    frame.resize(encoded_header.size() + context.stream_len);
+    std::memcpy(frame.data(), encoded_header.data(), encoded_header.size());
+
+    gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset,
+                                  frame.data() + encoded_header.size(), context.stream_len);
+
     return frame;
 }
 
@@ -202,7 +197,7 @@ void H264BitWriter::WriteBit(bool state) {
     WriteBits(state ? 1 : 0, 1);
 }
 
-void H264BitWriter::WriteScalingList(const std::vector<u8>& list, s32 start, s32 count) {
+void H264BitWriter::WriteScalingList(std::span<const u8> list, s32 start, s32 count) {
     std::vector<u8> scan(count);
     if (count == 16) {
         std::memcpy(scan.data(), zig_zag_scan.data(), scan.size());

src/video_core/command_classes/codecs/h264.h

@@ -20,7 +20,9 @@
 
 #pragma once
 
+#include <span>
 #include <vector>
+#include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "video_core/command_classes/nvdec_common.h"
@@ -48,7 +50,7 @@ public:
 
     /// Based on section 7.3.2.1.1.1 and Table 7-4 in the H.264 specification
     /// Writes the scaling matrices of the sream
-    void WriteScalingList(const std::vector<u8>& list, s32 start, s32 count);
+    void WriteScalingList(std::span<const u8> list, s32 start, s32 count);
 
     /// Return the bitstream as a vector.
     [[nodiscard]] std::vector<u8>& GetByteArray();
@@ -78,40 +80,110 @@ public:
         const NvdecCommon::NvdecRegisters& state, bool is_first_frame = false);
 
 private:
-    struct H264ParameterSet {
-        u32 log2_max_pic_order_cnt{};
-        u32 delta_pic_order_always_zero_flag{};
-        u32 frame_mbs_only_flag{};
-        u32 pic_width_in_mbs{};
-        u32 pic_height_in_map_units{};
-        INSERT_PADDING_WORDS(1);
-        u32 entropy_coding_mode_flag{};
-        u32 bottom_field_pic_order_flag{};
-        u32 num_refidx_l0_default_active{};
-        u32 num_refidx_l1_default_active{};
-        u32 deblocking_filter_control_flag{};
-        u32 redundant_pic_count_flag{};
-        u32 transform_8x8_mode_flag{};
-        INSERT_PADDING_WORDS(9);
-        u64 flags{};
-        u32 frame_number{};
-        u32 frame_number2{};
-    };
-    static_assert(sizeof(H264ParameterSet) == 0x68, "H264ParameterSet is an invalid size");
-
-    struct H264DecoderContext {
-        INSERT_PADDING_BYTES(0x48);
-        u32 frame_data_size{};
-        INSERT_PADDING_BYTES(0xc);
-        H264ParameterSet h264_parameter_set{};
-        INSERT_PADDING_BYTES(0x100);
-        std::array<u8, 0x60> scaling_matrix_4;
-        std::array<u8, 0x80> scaling_matrix_8;
-    };
-    static_assert(sizeof(H264DecoderContext) == 0x2a0, "H264DecoderContext is an invalid size");
-
     std::vector<u8> frame;
     GPU& gpu;
+
+    struct H264ParameterSet {
+        s32 log2_max_pic_order_cnt_lsb_minus4; ///< 0x00
+        s32 delta_pic_order_always_zero_flag;  ///< 0x04
+        s32 frame_mbs_only_flag;               ///< 0x08
+        u32 pic_width_in_mbs;                  ///< 0x0C
+        u32 frame_height_in_map_units;         ///< 0x10
+        union {                                ///< 0x14
+            BitField<0, 2, u32> tile_format;
+            BitField<2, 3, u32> gob_height;
+        };
+        u32 entropy_coding_mode_flag;               ///< 0x18
+        s32 pic_order_present_flag;                 ///< 0x1C
+        s32 num_refidx_l0_default_active;           ///< 0x20
+        s32 num_refidx_l1_default_active;           ///< 0x24
+        s32 deblocking_filter_control_present_flag; ///< 0x28
+        s32 redundant_pic_cnt_present_flag;         ///< 0x2C
+        u32 transform_8x8_mode_flag;                ///< 0x30
+        u32 pitch_luma;                             ///< 0x34
+        u32 pitch_chroma;                           ///< 0x38
+        u32 luma_top_offset;                        ///< 0x3C
+        u32 luma_bot_offset;                        ///< 0x40
+        u32 luma_frame_offset;                      ///< 0x44
+        u32 chroma_top_offset;                      ///< 0x48
+        u32 chroma_bot_offset;                      ///< 0x4C
+        u32 chroma_frame_offset;                    ///< 0x50
+        u32 hist_buffer_size;                       ///< 0x54
+        union {                                     ///< 0x58
+            union {
+                BitField<0, 1, u64> mbaff_frame;
+                BitField<1, 1, u64> direct_8x8_inference;
+                BitField<2, 1, u64> weighted_pred;
+                BitField<3, 1, u64> constrained_intra_pred;
+                BitField<4, 1, u64> ref_pic;
+                BitField<5, 1, u64> field_pic;
+                BitField<6, 1, u64> bottom_field;
+                BitField<7, 1, u64> second_field;
+            } flags;
+            BitField<8, 4, u64> log2_max_frame_num_minus4;
+            BitField<12, 2, u64> chroma_format_idc;
+            BitField<14, 2, u64> pic_order_cnt_type;
+            BitField<16, 6, s64> pic_init_qp_minus26;
+            BitField<22, 5, s64> chroma_qp_index_offset;
+            BitField<27, 5, s64> second_chroma_qp_index_offset;
+            BitField<32, 2, u64> weighted_bipred_idc;
+            BitField<34, 7, u64> curr_pic_idx;
+            BitField<41, 5, u64> curr_col_idx;
+            BitField<46, 16, u64> frame_number;
+            BitField<62, 1, u64> frame_surfaces;
+            BitField<63, 1, u64> output_memory_layout;
+        };
+    };
+    static_assert(sizeof(H264ParameterSet) == 0x60, "H264ParameterSet is an invalid size");
+
+    struct H264DecoderContext {
+        INSERT_PADDING_WORDS_NOINIT(18);       ///< 0x0000
+        u32 stream_len;                        ///< 0x0048
+        INSERT_PADDING_WORDS_NOINIT(3);        ///< 0x004C
+        H264ParameterSet h264_parameter_set;   ///< 0x0058
+        INSERT_PADDING_WORDS_NOINIT(66);       ///< 0x00B8
+        std::array<u8, 0x60> weight_scale;     ///< 0x01C0
+        std::array<u8, 0x80> weight_scale_8x8; ///< 0x0220
+    };
+    static_assert(sizeof(H264DecoderContext) == 0x2A0, "H264DecoderContext is an invalid size");
+
+#define ASSERT_POSITION(field_name, position)                                                      \
+    static_assert(offsetof(H264ParameterSet, field_name) == position,                              \
+                  "Field " #field_name " has invalid position")
+
+    ASSERT_POSITION(log2_max_pic_order_cnt_lsb_minus4, 0x00);
+    ASSERT_POSITION(delta_pic_order_always_zero_flag, 0x04);
+    ASSERT_POSITION(frame_mbs_only_flag, 0x08);
+    ASSERT_POSITION(pic_width_in_mbs, 0x0C);
+    ASSERT_POSITION(frame_height_in_map_units, 0x10);
+    ASSERT_POSITION(tile_format, 0x14);
+    ASSERT_POSITION(entropy_coding_mode_flag, 0x18);
+    ASSERT_POSITION(pic_order_present_flag, 0x1C);
+    ASSERT_POSITION(num_refidx_l0_default_active, 0x20);
+    ASSERT_POSITION(num_refidx_l1_default_active, 0x24);
+    ASSERT_POSITION(deblocking_filter_control_present_flag, 0x28);
+    ASSERT_POSITION(redundant_pic_cnt_present_flag, 0x2C);
+    ASSERT_POSITION(transform_8x8_mode_flag, 0x30);
+    ASSERT_POSITION(pitch_luma, 0x34);
+    ASSERT_POSITION(pitch_chroma, 0x38);
+    ASSERT_POSITION(luma_top_offset, 0x3C);
+    ASSERT_POSITION(luma_bot_offset, 0x40);
+    ASSERT_POSITION(luma_frame_offset, 0x44);
+    ASSERT_POSITION(chroma_top_offset, 0x48);
+    ASSERT_POSITION(chroma_bot_offset, 0x4C);
+    ASSERT_POSITION(chroma_frame_offset, 0x50);
+    ASSERT_POSITION(hist_buffer_size, 0x54);
+    ASSERT_POSITION(flags, 0x58);
+#undef ASSERT_POSITION
+
+#define ASSERT_POSITION(field_name, position)                                                      \
+    static_assert(offsetof(H264DecoderContext, field_name) == position,                            \
+                  "Field " #field_name " has invalid position")
+
+    ASSERT_POSITION(stream_len, 0x48);
+    ASSERT_POSITION(h264_parameter_set, 0x58);
+    ASSERT_POSITION(weight_scale, 0x1C0);
+#undef ASSERT_POSITION
 };
 
 } // namespace Decoder

src/video_core/command_classes/codecs/vp9.cpp

@@ -354,7 +354,7 @@ void VP9::WriteMvProbabilityUpdate(VpxRangeEncoder& writer, u8 new_prob, u8 old_
 }
 
 Vp9PictureInfo VP9::GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state) {
-    PictureInfo picture_info{};
+    PictureInfo picture_info;
     gpu.MemoryManager().ReadBlock(state.picture_info_offset, &picture_info, sizeof(PictureInfo));
     Vp9PictureInfo vp9_info = picture_info.Convert();
 
@@ -370,7 +370,7 @@ Vp9PictureInfo VP9::GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state)
 }
 
 void VP9::InsertEntropy(u64 offset, Vp9EntropyProbs& dst) {
-    EntropyProbs entropy{};
+    EntropyProbs entropy;
     gpu.MemoryManager().ReadBlock(offset, &entropy, sizeof(EntropyProbs));
     entropy.Convert(dst);
 }

src/video_core/command_classes/codecs/vp9_types.h

@@ -15,10 +15,10 @@ class GPU;
 
 namespace Decoder {
 struct Vp9FrameDimensions {
-    s16 width{};
-    s16 height{};
-    s16 luma_pitch{};
-    s16 chroma_pitch{};
+    s16 width;
+    s16 height;
+    s16 luma_pitch;
+    s16 chroma_pitch;
 };
 static_assert(sizeof(Vp9FrameDimensions) == 0x8, "Vp9 Vp9FrameDimensions is an invalid size");
 
@@ -49,87 +49,87 @@ enum class TxMode {
 };
 
 struct Segmentation {
-    u8 enabled{};
-    u8 update_map{};
-    u8 temporal_update{};
-    u8 abs_delta{};
-    std::array<u32, 8> feature_mask{};
-    std::array<std::array<s16, 4>, 8> feature_data{};
+    u8 enabled;
+    u8 update_map;
+    u8 temporal_update;
+    u8 abs_delta;
+    std::array<u32, 8> feature_mask;
+    std::array<std::array<s16, 4>, 8> feature_data;
 };
 static_assert(sizeof(Segmentation) == 0x64, "Segmentation is an invalid size");
 
 struct LoopFilter {
-    u8 mode_ref_delta_enabled{};
-    std::array<s8, 4> ref_deltas{};
-    std::array<s8, 2> mode_deltas{};
+    u8 mode_ref_delta_enabled;
+    std::array<s8, 4> ref_deltas;
+    std::array<s8, 2> mode_deltas;
 };
 static_assert(sizeof(LoopFilter) == 0x7, "LoopFilter is an invalid size");
 
 struct Vp9EntropyProbs {
-    std::array<u8, 36> y_mode_prob{};
-    std::array<u8, 64> partition_prob{};
-    std::array<u8, 1728> coef_probs{};
-    std::array<u8, 8> switchable_interp_prob{};
-    std::array<u8, 28> inter_mode_prob{};
-    std::array<u8, 4> intra_inter_prob{};
-    std::array<u8, 5> comp_inter_prob{};
-    std::array<u8, 10> single_ref_prob{};
-    std::array<u8, 5> comp_ref_prob{};
-    std::array<u8, 6> tx_32x32_prob{};
-    std::array<u8, 4> tx_16x16_prob{};
-    std::array<u8, 2> tx_8x8_prob{};
-    std::array<u8, 3> skip_probs{};
-    std::array<u8, 3> joints{};
-    std::array<u8, 2> sign{};
-    std::array<u8, 20> classes{};
-    std::array<u8, 2> class_0{};
-    std::array<u8, 20> prob_bits{};
-    std::array<u8, 12> class_0_fr{};
-    std::array<u8, 6> fr{};
-    std::array<u8, 2> class_0_hp{};
-    std::array<u8, 2> high_precision{};
+    std::array<u8, 36> y_mode_prob;           ///< 0x0000
+    std::array<u8, 64> partition_prob;        ///< 0x0024
+    std::array<u8, 1728> coef_probs;          ///< 0x0064
+    std::array<u8, 8> switchable_interp_prob; ///< 0x0724
+    std::array<u8, 28> inter_mode_prob;       ///< 0x072C
+    std::array<u8, 4> intra_inter_prob;       ///< 0x0748
+    std::array<u8, 5> comp_inter_prob;        ///< 0x074C
+    std::array<u8, 10> single_ref_prob;       ///< 0x0751
+    std::array<u8, 5> comp_ref_prob;          ///< 0x075B
+    std::array<u8, 6> tx_32x32_prob;          ///< 0x0760
+    std::array<u8, 4> tx_16x16_prob;          ///< 0x0766
+    std::array<u8, 2> tx_8x8_prob;            ///< 0x076A
+    std::array<u8, 3> skip_probs;             ///< 0x076C
+    std::array<u8, 3> joints;                 ///< 0x076F
+    std::array<u8, 2> sign;                   ///< 0x0772
+    std::array<u8, 20> classes;               ///< 0x0774
+    std::array<u8, 2> class_0;                ///< 0x0788
+    std::array<u8, 20> prob_bits;             ///< 0x078A
+    std::array<u8, 12> class_0_fr;            ///< 0x079E
+    std::array<u8, 6> fr;                     ///< 0x07AA
+    std::array<u8, 2> class_0_hp;             ///< 0x07B0
+    std::array<u8, 2> high_precision;         ///< 0x07B2
 };
 static_assert(sizeof(Vp9EntropyProbs) == 0x7B4, "Vp9EntropyProbs is an invalid size");
 
 struct Vp9PictureInfo {
-    bool is_key_frame{};
-    bool intra_only{};
-    bool last_frame_was_key{};
-    bool frame_size_changed{};
-    bool error_resilient_mode{};
-    bool last_frame_shown{};
-    bool show_frame{};
-    std::array<s8, 4> ref_frame_sign_bias{};
-    s32 base_q_index{};
-    s32 y_dc_delta_q{};
-    s32 uv_dc_delta_q{};
-    s32 uv_ac_delta_q{};
-    bool lossless{};
-    s32 transform_mode{};
-    bool allow_high_precision_mv{};
-    s32 interp_filter{};
-    s32 reference_mode{};
-    s8 comp_fixed_ref{};
-    std::array<s8, 2> comp_var_ref{};
-    s32 log2_tile_cols{};
-    s32 log2_tile_rows{};
-    bool segment_enabled{};
-    bool segment_map_update{};
-    bool segment_map_temporal_update{};
-    s32 segment_abs_delta{};
-    std::array<u32, 8> segment_feature_enable{};
-    std::array<std::array<s16, 4>, 8> segment_feature_data{};
-    bool mode_ref_delta_enabled{};
-    bool use_prev_in_find_mv_refs{};
-    std::array<s8, 4> ref_deltas{};
-    std::array<s8, 2> mode_deltas{};
-    Vp9EntropyProbs entropy{};
-    Vp9FrameDimensions frame_size{};
-    u8 first_level{};
-    u8 sharpness_level{};
-    u32 bitstream_size{};
-    std::array<u64, 4> frame_offsets{};
-    std::array<bool, 4> refresh_frame{};
+    bool is_key_frame;
+    bool intra_only;
+    bool last_frame_was_key;
+    bool frame_size_changed;
+    bool error_resilient_mode;
+    bool last_frame_shown;
+    bool show_frame;
+    std::array<s8, 4> ref_frame_sign_bias;
+    s32 base_q_index;
+    s32 y_dc_delta_q;
+    s32 uv_dc_delta_q;
+    s32 uv_ac_delta_q;
+    bool lossless;
+    s32 transform_mode;
+    bool allow_high_precision_mv;
+    s32 interp_filter;
+    s32 reference_mode;
+    s8 comp_fixed_ref;
+    std::array<s8, 2> comp_var_ref;
+    s32 log2_tile_cols;
+    s32 log2_tile_rows;
+    bool segment_enabled;
+    bool segment_map_update;
+    bool segment_map_temporal_update;
+    s32 segment_abs_delta;
+    std::array<u32, 8> segment_feature_enable;
+    std::array<std::array<s16, 4>, 8> segment_feature_data;
+    bool mode_ref_delta_enabled;
+    bool use_prev_in_find_mv_refs;
+    std::array<s8, 4> ref_deltas;
+    std::array<s8, 2> mode_deltas;
+    Vp9EntropyProbs entropy;
+    Vp9FrameDimensions frame_size;
+    u8 first_level;
+    u8 sharpness_level;
+    u32 bitstream_size;
+    std::array<u64, 4> frame_offsets;
+    std::array<bool, 4> refresh_frame;
 };
 
 struct Vp9FrameContainer {
@@ -138,35 +138,35 @@ struct Vp9FrameContainer {
 };
 
 struct PictureInfo {
-    INSERT_PADDING_WORDS(12);
-    u32 bitstream_size{};
-    INSERT_PADDING_WORDS(5);
-    Vp9FrameDimensions last_frame_size{};
-    Vp9FrameDimensions golden_frame_size{};
-    Vp9FrameDimensions alt_frame_size{};
-    Vp9FrameDimensions current_frame_size{};
-    u32 vp9_flags{};
-    std::array<s8, 4> ref_frame_sign_bias{};
-    u8 first_level{};
-    u8 sharpness_level{};
-    u8 base_q_index{};
-    u8 y_dc_delta_q{};
-    u8 uv_ac_delta_q{};
-    u8 uv_dc_delta_q{};
-    u8 lossless{};
-    u8 tx_mode{};
-    u8 allow_high_precision_mv{};
-    u8 interp_filter{};
-    u8 reference_mode{};
-    s8 comp_fixed_ref{};
-    std::array<s8, 2> comp_var_ref{};
-    u8 log2_tile_cols{};
-    u8 log2_tile_rows{};
-    Segmentation segmentation{};
-    LoopFilter loop_filter{};
-    INSERT_PADDING_BYTES(5);
-    u32 surface_params{};
-    INSERT_PADDING_WORDS(3);
+    INSERT_PADDING_WORDS_NOINIT(12);       ///< 0x00
+    u32 bitstream_size;                    ///< 0x30
+    INSERT_PADDING_WORDS_NOINIT(5);        ///< 0x34
+    Vp9FrameDimensions last_frame_size;    ///< 0x48
+    Vp9FrameDimensions golden_frame_size;  ///< 0x50
+    Vp9FrameDimensions alt_frame_size;     ///< 0x58
+    Vp9FrameDimensions current_frame_size; ///< 0x60
+    u32 vp9_flags;                         ///< 0x68
+    std::array<s8, 4> ref_frame_sign_bias; ///< 0x6C
+    u8 first_level;                        ///< 0x70
+    u8 sharpness_level;                    ///< 0x71
+    u8 base_q_index;                       ///< 0x72
+    u8 y_dc_delta_q;                       ///< 0x73
+    u8 uv_ac_delta_q;                      ///< 0x74
+    u8 uv_dc_delta_q;                      ///< 0x75
+    u8 lossless;                           ///< 0x76
+    u8 tx_mode;                            ///< 0x77
+    u8 allow_high_precision_mv;            ///< 0x78
+    u8 interp_filter;                      ///< 0x79
+    u8 reference_mode;                     ///< 0x7A
+    s8 comp_fixed_ref;                     ///< 0x7B
+    std::array<s8, 2> comp_var_ref;        ///< 0x7C
+    u8 log2_tile_cols;                     ///< 0x7E
+    u8 log2_tile_rows;                     ///< 0x7F
+    Segmentation segmentation;             ///< 0x80
+    LoopFilter loop_filter;                ///< 0xE4
+    INSERT_PADDING_BYTES_NOINIT(5);        ///< 0xEB
+    u32 surface_params;                    ///< 0xF0
+    INSERT_PADDING_WORDS_NOINIT(3);        ///< 0xF4
 
     [[nodiscard]] Vp9PictureInfo Convert() const {
         return {
@@ -176,6 +176,7 @@ struct PictureInfo {
             .frame_size_changed = (vp9_flags & FrameFlags::FrameSizeChanged) != 0,
             .error_resilient_mode = (vp9_flags & FrameFlags::ErrorResilientMode) != 0,
             .last_frame_shown = (vp9_flags & FrameFlags::LastShowFrame) != 0,
+            .show_frame = false,
             .ref_frame_sign_bias = ref_frame_sign_bias,
             .base_q_index = base_q_index,
             .y_dc_delta_q = y_dc_delta_q,
@@ -204,45 +205,48 @@ struct PictureInfo {
                                         !(vp9_flags == (FrameFlags::LastFrameIsKeyFrame)),
             .ref_deltas = loop_filter.ref_deltas,
             .mode_deltas = loop_filter.mode_deltas,
+            .entropy{},
             .frame_size = current_frame_size,
             .first_level = first_level,
             .sharpness_level = sharpness_level,
             .bitstream_size = bitstream_size,
+            .frame_offsets{},
+            .refresh_frame{},
         };
     }
 };
 static_assert(sizeof(PictureInfo) == 0x100, "PictureInfo is an invalid size");
 
 struct EntropyProbs {
-    INSERT_PADDING_BYTES(1024);
-    std::array<u8, 28> inter_mode_prob{};
-    std::array<u8, 4> intra_inter_prob{};
-    INSERT_PADDING_BYTES(80);
-    std::array<u8, 2> tx_8x8_prob{};
-    std::array<u8, 4> tx_16x16_prob{};
-    std::array<u8, 6> tx_32x32_prob{};
-    std::array<u8, 4> y_mode_prob_e8{};
-    std::array<std::array<u8, 8>, 4> y_mode_prob_e0e7{};
-    INSERT_PADDING_BYTES(64);
-    std::array<u8, 64> partition_prob{};
-    INSERT_PADDING_BYTES(10);
-    std::array<u8, 8> switchable_interp_prob{};
-    std::array<u8, 5> comp_inter_prob{};
-    std::array<u8, 3> skip_probs{};
-    INSERT_PADDING_BYTES(1);
-    std::array<u8, 3> joints{};
-    std::array<u8, 2> sign{};
-    std::array<u8, 2> class_0{};
-    std::array<u8, 6> fr{};
-    std::array<u8, 2> class_0_hp{};
-    std::array<u8, 2> high_precision{};
-    std::array<u8, 20> classes{};
-    std::array<u8, 12> class_0_fr{};
-    std::array<u8, 20> pred_bits{};
-    std::array<u8, 10> single_ref_prob{};
-    std::array<u8, 5> comp_ref_prob{};
-    INSERT_PADDING_BYTES(17);
-    std::array<u8, 2304> coef_probs{};
+    INSERT_PADDING_BYTES_NOINIT(1024);                 ///< 0x0000
+    std::array<u8, 28> inter_mode_prob;                ///< 0x0400
+    std::array<u8, 4> intra_inter_prob;                ///< 0x041C
+    INSERT_PADDING_BYTES_NOINIT(80);                   ///< 0x0420
+    std::array<u8, 2> tx_8x8_prob;                     ///< 0x0470
+    std::array<u8, 4> tx_16x16_prob;                   ///< 0x0472
+    std::array<u8, 6> tx_32x32_prob;                   ///< 0x0476
+    std::array<u8, 4> y_mode_prob_e8;                  ///< 0x047C
+    std::array<std::array<u8, 8>, 4> y_mode_prob_e0e7; ///< 0x0480
+    INSERT_PADDING_BYTES_NOINIT(64);                   ///< 0x04A0
+    std::array<u8, 64> partition_prob;                 ///< 0x04E0
+    INSERT_PADDING_BYTES_NOINIT(10);                   ///< 0x0520
+    std::array<u8, 8> switchable_interp_prob;          ///< 0x052A
+    std::array<u8, 5> comp_inter_prob;                 ///< 0x0532
+    std::array<u8, 3> skip_probs;                      ///< 0x0537
+    INSERT_PADDING_BYTES_NOINIT(1);                    ///< 0x053A
+    std::array<u8, 3> joints;                          ///< 0x053B
+    std::array<u8, 2> sign;                            ///< 0x053E
+    std::array<u8, 2> class_0;                         ///< 0x0540
+    std::array<u8, 6> fr;                              ///< 0x0542
+    std::array<u8, 2> class_0_hp;                      ///< 0x0548
+    std::array<u8, 2> high_precision;                  ///< 0x054A
+    std::array<u8, 20> classes;                        ///< 0x054C
+    std::array<u8, 12> class_0_fr;                     ///< 0x0560
+    std::array<u8, 20> pred_bits;                      ///< 0x056C
+    std::array<u8, 10> single_ref_prob;                ///< 0x0580
+    std::array<u8, 5> comp_ref_prob;                   ///< 0x058A
+    INSERT_PADDING_BYTES_NOINIT(17);                   ///< 0x058F
+    std::array<u8, 2304> coef_probs;                   ///< 0x05A0
 
     void Convert(Vp9EntropyProbs& fc) {
         fc.inter_mode_prob = inter_mode_prob;
@@ -293,10 +297,45 @@ struct RefPoolElement {
 };
 
 struct FrameContexts {
-    s64 from{};
-    bool adapted{};
-    Vp9EntropyProbs probs{};
+    s64 from;
+    bool adapted;
+    Vp9EntropyProbs probs;
 };
 
+#define ASSERT_POSITION(field_name, position)                                                      \
+    static_assert(offsetof(Vp9EntropyProbs, field_name) == position,                               \
+                  "Field " #field_name " has invalid position")
+
+ASSERT_POSITION(partition_prob, 0x0024);
+ASSERT_POSITION(switchable_interp_prob, 0x0724);
+ASSERT_POSITION(sign, 0x0772);
+ASSERT_POSITION(class_0_fr, 0x079E);
+ASSERT_POSITION(high_precision, 0x07B2);
+#undef ASSERT_POSITION
+
+#define ASSERT_POSITION(field_name, position)                                                      \
+    static_assert(offsetof(PictureInfo, field_name) == position,                                   \
+                  "Field " #field_name " has invalid position")
+
+ASSERT_POSITION(bitstream_size, 0x30);
+ASSERT_POSITION(last_frame_size, 0x48);
+ASSERT_POSITION(first_level, 0x70);
+ASSERT_POSITION(segmentation, 0x80);
+ASSERT_POSITION(loop_filter, 0xE4);
+ASSERT_POSITION(surface_params, 0xF0);
+#undef ASSERT_POSITION
+
+#define ASSERT_POSITION(field_name, position)                                                      \
+    static_assert(offsetof(EntropyProbs, field_name) == position,                                  \
+                  "Field " #field_name " has invalid position")
+
+ASSERT_POSITION(inter_mode_prob, 0x400);
+ASSERT_POSITION(tx_8x8_prob, 0x470);
+ASSERT_POSITION(partition_prob, 0x4E0);
+ASSERT_POSITION(class_0, 0x540);
+ASSERT_POSITION(class_0_fr, 0x560);
+ASSERT_POSITION(coef_probs, 0x5A0);
+#undef ASSERT_POSITION
+
 }; // namespace Decoder
 }; // namespace Tegra

src/video_core/command_classes/nvdec.cpp

@@ -8,22 +8,21 @@
 
 namespace Tegra {
 
-Nvdec::Nvdec(GPU& gpu_) : gpu(gpu_), codec(std::make_unique<Codec>(gpu)) {}
+#define NVDEC_REG_INDEX(field_name)                                                                \
+    (offsetof(NvdecCommon::NvdecRegisters, field_name) / sizeof(u64))
+
+Nvdec::Nvdec(GPU& gpu_) : gpu(gpu_), state{}, codec(std::make_unique<Codec>(gpu, state)) {}
 
 Nvdec::~Nvdec() = default;
 
-void Nvdec::ProcessMethod(Method method, u32 argument) {
-    if (method == Method::SetVideoCodec) {
-        codec->StateWrite(static_cast<u32>(method), argument);
-    } else {
-        codec->StateWrite(static_cast<u32>(method), static_cast<u64>(argument) << 8);
-    }
+void Nvdec::ProcessMethod(u32 method, u32 argument) {
+    state.reg_array[method] = static_cast<u64>(argument) << 8;
 
     switch (method) {
-    case Method::SetVideoCodec:
+    case NVDEC_REG_INDEX(set_codec_id):
         codec->SetTargetCodec(static_cast<NvdecCommon::VideoCodec>(argument));
         break;
-    case Method::Execute:
+    case NVDEC_REG_INDEX(execute):
         Execute();
         break;
     }

src/video_core/command_classes/nvdec.h

@@ -14,16 +14,11 @@ class GPU;
 
 class Nvdec {
 public:
-    enum class Method : u32 {
-        SetVideoCodec = 0x80,
-        Execute = 0xc0,
-    };
-
     explicit Nvdec(GPU& gpu);
     ~Nvdec();
 
     /// Writes the method into the state, Invoke Execute() if encountered
-    void ProcessMethod(Method method, u32 argument);
+    void ProcessMethod(u32 method, u32 argument);
 
     /// Return most recently decoded frame
     [[nodiscard]] AVFramePtr GetFrame();
@@ -33,6 +28,7 @@ private:
     void Execute();
 
     GPU& gpu;
+    NvdecCommon::NvdecRegisters state;
     std::unique_ptr<Codec> codec;
 };
 } // namespace Tegra

src/video_core/command_classes/nvdec_common.h

@@ -4,40 +4,13 @@
 
 #pragma once
 
+#include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 
 namespace Tegra::NvdecCommon {
 
-struct NvdecRegisters {
-    INSERT_PADDING_WORDS(256);
-    u64 set_codec_id{};
-    INSERT_PADDING_WORDS(254);
-    u64 set_platform_id{};
-    u64 picture_info_offset{};
-    u64 frame_bitstream_offset{};
-    u64 frame_number{};
-    u64 h264_slice_data_offsets{};
-    u64 h264_mv_dump_offset{};
-    INSERT_PADDING_WORDS(6);
-    u64 frame_stats_offset{};
-    u64 h264_last_surface_luma_offset{};
-    u64 h264_last_surface_chroma_offset{};
-    std::array<u64, 17> surface_luma_offset{};
-    std::array<u64, 17> surface_chroma_offset{};
-    INSERT_PADDING_WORDS(132);
-    u64 vp9_entropy_probs_offset{};
-    u64 vp9_backward_updates_offset{};
-    u64 vp9_last_frame_segmap_offset{};
-    u64 vp9_curr_frame_segmap_offset{};
-    INSERT_PADDING_WORDS(2);
-    u64 vp9_last_frame_mvs_offset{};
-    u64 vp9_curr_frame_mvs_offset{};
-    INSERT_PADDING_WORDS(2);
-};
-static_assert(sizeof(NvdecRegisters) == (0xBC0), "NvdecRegisters is incorrect size");
-
-enum class VideoCodec : u32 {
+enum class VideoCodec : u64 {
     None = 0x0,
     H264 = 0x3,
     Vp8 = 0x5,
@@ -45,4 +18,76 @@ enum class VideoCodec : u32 {
     Vp9 = 0x9,
 };
 
+// NVDEC should use a 32-bit address space, but is mapped to 64-bit,
+// doubling the sizes here is compensating for that.
+struct NvdecRegisters {
+    static constexpr std::size_t NUM_REGS = 0x178;
+
+    union {
+        struct {
+            INSERT_PADDING_WORDS_NOINIT(256); ///< 0x0000
+            VideoCodec set_codec_id;          ///< 0x0400
+            INSERT_PADDING_WORDS_NOINIT(126); ///< 0x0408
+            u64 execute;                      ///< 0x0600
+            INSERT_PADDING_WORDS_NOINIT(126); ///< 0x0608
+            struct {                          ///< 0x0800
+                union {
+                    BitField<0, 3, VideoCodec> codec;
+                    BitField<4, 1, u64> gp_timer_on;
+                    BitField<13, 1, u64> mb_timer_on;
+                    BitField<14, 1, u64> intra_frame_pslc;
+                    BitField<17, 1, u64> all_intra_frame;
+                };
+            } control_params;
+            u64 picture_info_offset;                   ///< 0x0808
+            u64 frame_bitstream_offset;                ///< 0x0810
+            u64 frame_number;                          ///< 0x0818
+            u64 h264_slice_data_offsets;               ///< 0x0820
+            u64 h264_mv_dump_offset;                   ///< 0x0828
+            INSERT_PADDING_WORDS_NOINIT(6);            ///< 0x0830
+            u64 frame_stats_offset;                    ///< 0x0848
+            u64 h264_last_surface_luma_offset;         ///< 0x0850
+            u64 h264_last_surface_chroma_offset;       ///< 0x0858
+            std::array<u64, 17> surface_luma_offset;   ///< 0x0860
+            std::array<u64, 17> surface_chroma_offset; ///< 0x08E8
+            INSERT_PADDING_WORDS_NOINIT(132);          ///< 0x0970
+            u64 vp9_entropy_probs_offset;              ///< 0x0B80
+            u64 vp9_backward_updates_offset;           ///< 0x0B88
+            u64 vp9_last_frame_segmap_offset;          ///< 0x0B90
+            u64 vp9_curr_frame_segmap_offset;          ///< 0x0B98
+            INSERT_PADDING_WORDS_NOINIT(2);            ///< 0x0BA0
+            u64 vp9_last_frame_mvs_offset;             ///< 0x0BA8
+            u64 vp9_curr_frame_mvs_offset;             ///< 0x0BB0
+            INSERT_PADDING_WORDS_NOINIT(2);            ///< 0x0BB8
+        };
+        std::array<u64, NUM_REGS> reg_array;
+    };
+};
+static_assert(sizeof(NvdecRegisters) == (0xBC0), "NvdecRegisters is incorrect size");
+
+#define ASSERT_REG_POSITION(field_name, position)                                                  \
+    static_assert(offsetof(NvdecRegisters, field_name) == position * sizeof(u64),                  \
+                  "Field " #field_name " has invalid position")
+
+ASSERT_REG_POSITION(set_codec_id, 0x80);
+ASSERT_REG_POSITION(execute, 0xC0);
+ASSERT_REG_POSITION(control_params, 0x100);
+ASSERT_REG_POSITION(picture_info_offset, 0x101);
+ASSERT_REG_POSITION(frame_bitstream_offset, 0x102);
+ASSERT_REG_POSITION(frame_number, 0x103);
+ASSERT_REG_POSITION(h264_slice_data_offsets, 0x104);
+ASSERT_REG_POSITION(frame_stats_offset, 0x109);
+ASSERT_REG_POSITION(h264_last_surface_luma_offset, 0x10A);
+ASSERT_REG_POSITION(h264_last_surface_chroma_offset, 0x10B);
+ASSERT_REG_POSITION(surface_luma_offset, 0x10C);
+ASSERT_REG_POSITION(surface_chroma_offset, 0x11D);
+ASSERT_REG_POSITION(vp9_entropy_probs_offset, 0x170);
+ASSERT_REG_POSITION(vp9_backward_updates_offset, 0x171);
+ASSERT_REG_POSITION(vp9_last_frame_segmap_offset, 0x172);
+ASSERT_REG_POSITION(vp9_curr_frame_segmap_offset, 0x173);
+ASSERT_REG_POSITION(vp9_last_frame_mvs_offset, 0x175);
+ASSERT_REG_POSITION(vp9_curr_frame_mvs_offset, 0x176);
+
+#undef ASSERT_REG_POSITION
+
 } // namespace Tegra::NvdecCommon

src/video_core/command_classes/vic.cpp

@@ -9,17 +9,10 @@ extern "C" {
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wconversion"
 #endif
-#ifdef _MSC_VER
-#pragma warning(disable : 4244) // conversion from 'type' to 'type', possible loss of data
-#pragma warning(push)
-#endif
 #include <libswscale/swscale.h>
 #if defined(__GNUC__) || defined(__clang__)
 #pragma GCC diagnostic pop
 #endif
-#ifdef _MSC_VER
-#pragma warning(pop)
-#endif
 }
 
 #include "common/assert.h"

src/video_core/engines/shader_bytecode.h

@@ -1785,6 +1785,8 @@ public:
         SSY,
         SYNC,
         BRK,
+        CAL,
+        RET,
         DEPBAR,
         VOTE,
         VOTE_VTG,
@@ -2108,6 +2110,8 @@ private:
             INST("1111000011111---", Id::SYNC, Type::Flow, "SYNC"),
             INST("111000110100----", Id::BRK, Type::Flow, "BRK"),
             INST("111000110000----", Id::EXIT, Type::Flow, "EXIT"),
+            INST("111000100110----", Id::CAL, Type::Flow, "CAL"),
+            INST("111000110010----", Id::RET, Type::Flow, "RET"),
             INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"),
             INST("0101000011011---", Id::VOTE, Type::Warp, "VOTE"),
             INST("0101000011100---", Id::VOTE_VTG, Type::Warp, "VOTE_VTG"),

src/video_core/memory_manager.cpp

@@ -69,11 +69,16 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
     } else {
         UNREACHABLE_MSG("Unmapping non-existent GPU address=0x{:x}", gpu_addr);
     }
-    // Flush and invalidate through the GPU interface, to be asynchronous if possible.
-    const std::optional<VAddr> cpu_addr = GpuToCpuAddress(gpu_addr);
-    ASSERT(cpu_addr);
 
-    rasterizer->UnmapMemory(*cpu_addr, size);
+    const auto submapped_ranges = GetSubmappedRange(gpu_addr, size);
+
+    for (const auto& map : submapped_ranges) {
+        // Flush and invalidate through the GPU interface, to be asynchronous if possible.
+        const std::optional<VAddr> cpu_addr = GpuToCpuAddress(map.first);
+        ASSERT(cpu_addr);
+
+        rasterizer->UnmapMemory(*cpu_addr, map.second);
+    }
 
     UpdateRange(gpu_addr, PageEntry::State::Unmapped, size);
 }
@@ -114,6 +119,25 @@ void MemoryManager::TryUnlockPage(PageEntry page_entry, std::size_t size) {
                .IsSuccess());
 }
 
+void MemoryManager::UnmapVicFrame(GPUVAddr gpu_addr, std::size_t size) {
+    if (!size) {
+        return;
+    }
+
+    const std::optional<VAddr> cpu_addr = GpuToCpuAddress(gpu_addr);
+    ASSERT(cpu_addr);
+    rasterizer->InvalidateExceptTextureCache(*cpu_addr, size);
+    cache_invalidate_queue.push_back({*cpu_addr, size});
+
+    UpdateRange(gpu_addr, PageEntry::State::Unmapped, size);
+}
+
+void MemoryManager::InvalidateQueuedCaches() {
+    for (const auto& entry : cache_invalidate_queue) {
+        rasterizer->InvalidateTextureCache(entry.first, entry.second);
+    }
+    cache_invalidate_queue.clear();
+}
 PageEntry MemoryManager::GetPageEntry(GPUVAddr gpu_addr) const {
     return page_table[PageEntryIndex(gpu_addr)];
 }
@@ -127,8 +151,14 @@ void MemoryManager::SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::s
 
     //// Lock the new page
     // TryLockPage(page_entry, size);
+    auto& current_page = page_table[PageEntryIndex(gpu_addr)];
 
-    page_table[PageEntryIndex(gpu_addr)] = page_entry;
+    if ((!current_page.IsValid() && page_entry.IsValid()) ||
+        current_page.ToAddress() != page_entry.ToAddress()) {
+        rasterizer->ModifyGPUMemory(gpu_addr, size);
+    }
+
+    current_page = page_entry;
 }
 
 std::optional<GPUVAddr> MemoryManager::FindFreeRange(std::size_t size, std::size_t align,
@@ -174,6 +204,19 @@ std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) const {
     return page_entry.ToAddress() + (gpu_addr & page_mask);
 }
 
+std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr, std::size_t size) const {
+    size_t page_index{addr >> page_bits};
+    const size_t page_last{(addr + size + page_size - 1) >> page_bits};
+    while (page_index < page_last) {
+        const auto page_addr{GpuToCpuAddress(page_index << page_bits)};
+        if (page_addr && *page_addr != 0) {
+            return page_addr;
+        }
+        ++page_index;
+    }
+    return std::nullopt;
+}
+
 template <typename T>
 T MemoryManager::Read(GPUVAddr addr) const {
     if (auto page_pointer{GetPointer(addr)}; page_pointer) {
@@ -370,4 +413,79 @@ bool MemoryManager::IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const {
     return page <= Core::Memory::PAGE_SIZE;
 }
 
+bool MemoryManager::IsContinousRange(GPUVAddr gpu_addr, std::size_t size) const {
+    size_t page_index{gpu_addr >> page_bits};
+    const size_t page_last{(gpu_addr + size + page_size - 1) >> page_bits};
+    std::optional<VAddr> old_page_addr{};
+    while (page_index != page_last) {
+        const auto page_addr{GpuToCpuAddress(page_index << page_bits)};
+        if (!page_addr || *page_addr == 0) {
+            return false;
+        }
+        if (old_page_addr) {
+            if (*old_page_addr + page_size != *page_addr) {
+                return false;
+            }
+        }
+        old_page_addr = page_addr;
+        ++page_index;
+    }
+    return true;
+}
+
+bool MemoryManager::IsFullyMappedRange(GPUVAddr gpu_addr, std::size_t size) const {
+    size_t page_index{gpu_addr >> page_bits};
+    const size_t page_last{(gpu_addr + size + page_size - 1) >> page_bits};
+    while (page_index < page_last) {
+        if (!page_table[page_index].IsValid() || page_table[page_index].ToAddress() == 0) {
+            return false;
+        }
+        ++page_index;
+    }
+    return true;
+}
+
+std::vector<std::pair<GPUVAddr, std::size_t>> MemoryManager::GetSubmappedRange(
+    GPUVAddr gpu_addr, std::size_t size) const {
+    std::vector<std::pair<GPUVAddr, std::size_t>> result{};
+    size_t page_index{gpu_addr >> page_bits};
+    size_t remaining_size{size};
+    size_t page_offset{gpu_addr & page_mask};
+    std::optional<std::pair<GPUVAddr, std::size_t>> last_segment{};
+    std::optional<VAddr> old_page_addr{};
+    const auto extend_size = [this, &last_segment, &page_index](std::size_t bytes) {
+        if (!last_segment) {
+            GPUVAddr new_base_addr = page_index << page_bits;
+            last_segment = {new_base_addr, bytes};
+        } else {
+            last_segment->second += bytes;
+        }
+    };
+    const auto split = [this, &last_segment, &result] {
+        if (last_segment) {
+            result.push_back(*last_segment);
+            last_segment = std::nullopt;
+        }
+    };
+    while (remaining_size > 0) {
+        const size_t num_bytes{std::min(page_size - page_offset, remaining_size)};
+        const auto page_addr{GpuToCpuAddress(page_index << page_bits)};
+        if (!page_addr) {
+            split();
+        } else if (old_page_addr) {
+            if (*old_page_addr + page_size != *page_addr) {
+                split();
+            }
+            extend_size(num_bytes);
+        } else {
+            extend_size(num_bytes);
+        }
+        ++page_index;
+        page_offset = 0;
+        remaining_size -= num_bytes;
+    }
+    split();
+    return result;
+}
+
 } // namespace Tegra

src/video_core/memory_manager.h

@@ -76,6 +76,8 @@ public:
 
     [[nodiscard]] std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr) const;
 
+    [[nodiscard]] std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr, std::size_t size) const;
+
     template <typename T>
     [[nodiscard]] T Read(GPUVAddr addr) const;
 
@@ -112,10 +114,28 @@ public:
     void WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size);
 
     /**
-     * IsGranularRange checks if a gpu region can be simply read with a pointer.
+     * Checks if a gpu region can be simply read with a pointer.
      */
     [[nodiscard]] bool IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const;
 
+    /**
+     * Checks if a gpu region is mapped by a single range of cpu addresses.
+     */
+    [[nodiscard]] bool IsContinousRange(GPUVAddr gpu_addr, std::size_t size) const;
+
+    /**
+     * Checks if a gpu region is mapped entirely.
+     */
+    [[nodiscard]] bool IsFullyMappedRange(GPUVAddr gpu_addr, std::size_t size) const;
+
+    /**
+     * Returns a vector with all the subranges of cpu addresses mapped beneath.
+     * if the region is continous, a single pair will be returned. If it's unmapped, an empty vector
+     * will be returned;
+     */
+    std::vector<std::pair<GPUVAddr, std::size_t>> GetSubmappedRange(GPUVAddr gpu_addr,
+                                                                    std::size_t size) const;
+
     [[nodiscard]] GPUVAddr Map(VAddr cpu_addr, GPUVAddr gpu_addr, std::size_t size);
     [[nodiscard]] GPUVAddr MapAllocate(VAddr cpu_addr, std::size_t size, std::size_t align);
     [[nodiscard]] GPUVAddr MapAllocate32(VAddr cpu_addr, std::size_t size);
@@ -123,6 +143,14 @@ public:
     [[nodiscard]] GPUVAddr Allocate(std::size_t size, std::size_t align);
     void Unmap(GPUVAddr gpu_addr, std::size_t size);
 
+    /**
+     * Some Decoded NVDEC frames require that texture cache does not get invalidated.
+     * UnmapVicFrame defers the texture cache invalidation until the stream ends
+     * by invoking InvalidateQueuedCaches to invalidate all frame texture caches.
+     */
+    void UnmapVicFrame(GPUVAddr gpu_addr, std::size_t size);
+    void InvalidateQueuedCaches();
+
 private:
     [[nodiscard]] PageEntry GetPageEntry(GPUVAddr gpu_addr) const;
     void SetPageEntry(GPUVAddr gpu_addr, PageEntry page_entry, std::size_t size = page_size);

src/video_core/rasterizer_interface.h

@@ -72,6 +72,12 @@ public:
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     virtual void FlushRegion(VAddr addr, u64 size) = 0;
 
+    /// Notify rasterizer to flush the texture cache to Switch memory
+    virtual void InvalidateExceptTextureCache(VAddr addr, u64 size) = 0;
+
+    /// Notify rasterizer to invalidate the texture cache
+    virtual void InvalidateTextureCache(VAddr addr, u64 size) = 0;
+
     /// Check if the the specified memory area requires flushing to CPU Memory.
     virtual bool MustFlushRegion(VAddr addr, u64 size) = 0;
 
@@ -87,6 +93,9 @@ public:
     /// Unmap memory range
     virtual void UnmapMemory(VAddr addr, u64 size) = 0;
 
+    /// Remap GPU memory range. This means underneath backing memory changed
+    virtual void ModifyGPUMemory(GPUVAddr addr, u64 size) = 0;
+
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     /// and invalidated
     virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;

src/video_core/renderer_opengl/gl_arb_decompiler.cpp

@@ -491,6 +491,9 @@ private:
     const Registry& registry;
     const ShaderType stage;
 
+    std::shared_ptr<ShaderFunctionIR> context_func;
+    u32 ast_var_base{};
+
     std::size_t num_temporaries = 0;
     std::size_t max_temporaries = 0;
 
@@ -807,13 +810,33 @@ ARBDecompiler::ARBDecompiler(const Device& device_, const ShaderIR& ir_, const R
     : device{device_}, ir{ir_}, registry{registry_}, stage{stage_} {
     DefineGlobalMemory();
 
+    context_func = ir.GetMainFunction();
+    ast_var_base = 0;
+
     AddLine("TEMP RC;");
     AddLine("TEMP FSWZA[4];");
     AddLine("TEMP FSWZB[4];");
-    if (ir.IsDecompiled()) {
+    InitializeVariables();
+    AddLine("main:");
+    if (context_func->IsDecompiled()) {
         DecompileAST();
     } else {
         DecompileBranchMode();
+        AddLine("RET;");
+    }
+
+    const auto& subfunctions = ir.GetSubFunctions();
+    auto it = subfunctions.begin();
+    while (it != subfunctions.end()) {
+        context_func = *it;
+        AddLine("func_{}:", context_func->GetId());
+        if (context_func->IsDecompiled()) {
+            DecompileAST();
+        } else {
+            DecompileBranchMode();
+            AddLine("RET;");
+        }
+        it++;
     }
     AddLine("END");
 
@@ -1060,41 +1083,38 @@ void ARBDecompiler::InitializeVariables() {
 }
 
 void ARBDecompiler::DecompileAST() {
-    const u32 num_flow_variables = ir.GetASTNumVariables();
+    const u32 num_flow_variables = context_func->GetASTNumVariables();
     for (u32 i = 0; i < num_flow_variables; ++i) {
-        AddLine("TEMP F{};", i);
+        AddLine("TEMP F{};", i + ast_var_base);
     }
     for (u32 i = 0; i < num_flow_variables; ++i) {
-        AddLine("MOV.U F{}, {{0, 0, 0, 0}};", i);
+        AddLine("MOV.U F{}, {{0, 0, 0, 0}};", i + ast_var_base);
     }
 
-    InitializeVariables();
-
-    VisitAST(ir.GetASTProgram());
+    VisitAST(context_func->GetASTProgram());
+    ast_var_base += num_flow_variables;
 }
 
 void ARBDecompiler::DecompileBranchMode() {
     static constexpr u32 FLOW_STACK_SIZE = 20;
-    if (!ir.IsFlowStackDisabled()) {
+    if (!context_func->IsFlowStackDisabled()) {
         AddLine("TEMP SSY[{}];", FLOW_STACK_SIZE);
         AddLine("TEMP PBK[{}];", FLOW_STACK_SIZE);
         AddLine("TEMP SSY_TOP;");
         AddLine("TEMP PBK_TOP;");
     }
 
-    AddLine("TEMP PC;");
+    AddLine("TEMP PC{};", context_func->GetId());
 
-    if (!ir.IsFlowStackDisabled()) {
+    if (!context_func->IsFlowStackDisabled()) {
         AddLine("MOV.U SSY_TOP.x, 0;");
         AddLine("MOV.U PBK_TOP.x, 0;");
     }
 
-    InitializeVariables();
-
-    const auto basic_block_end = ir.GetBasicBlocks().end();
-    auto basic_block_it = ir.GetBasicBlocks().begin();
+    const auto basic_block_end = context_func->GetBasicBlocks().end();
+    auto basic_block_it = context_func->GetBasicBlocks().begin();
     const u32 first_address = basic_block_it->first;
-    AddLine("MOV.U PC.x, {};", first_address);
+    AddLine("MOV.U PC{}.x, {};", context_func->GetId(), first_address);
 
     AddLine("REP;");
 
@@ -1103,7 +1123,7 @@ void ARBDecompiler::DecompileBranchMode() {
         const auto& [address, bb] = *basic_block_it;
         ++num_blocks;
 
-        AddLine("SEQ.S.CC RC.x, PC.x, {};", address);
+        AddLine("SEQ.S.CC RC.x, PC{}.x, {};", context_func->GetId(), address);
         AddLine("IF NE.x;");
 
         VisitBlock(bb);
@@ -1114,7 +1134,7 @@ void ARBDecompiler::DecompileBranchMode() {
             const auto op = std::get_if<OperationNode>(&*bb[bb.size() - 1]);
             if (!op || op->GetCode() != OperationCode::Branch) {
                 const u32 next_address = basic_block_it->first;
-                AddLine("MOV.U PC.x, {};", next_address);
+                AddLine("MOV.U PC{}.x, {};", context_func->GetId(), next_address);
                 AddLine("CONT;");
             }
         }
@@ -1152,7 +1172,8 @@ void ARBDecompiler::VisitAST(const ASTNode& node) {
     } else if (const auto decoded = std::get_if<ASTBlockDecoded>(&*node->GetInnerData())) {
         VisitBlock(decoded->nodes);
     } else if (const auto var_set = std::get_if<ASTVarSet>(&*node->GetInnerData())) {
-        AddLine("MOV.U F{}, {};", var_set->index, VisitExpression(var_set->condition));
+        AddLine("MOV.U F{}, {};", var_set->index + ast_var_base,
+                VisitExpression(var_set->condition));
         ResetTemporaries();
     } else if (const auto do_while = std::get_if<ASTDoWhile>(&*node->GetInnerData())) {
         const std::string condition = VisitExpression(do_while->condition);
@@ -1172,7 +1193,11 @@ void ARBDecompiler::VisitAST(const ASTNode& node) {
             ResetTemporaries();
         }
         if (ast_return->kills) {
-            AddLine("KIL TR;");
+            if (stage == ShaderType::Fragment) {
+                AddLine("KIL TR;");
+            } else {
+                AddLine("RET;");
+            }
         } else {
             Exit();
         }
@@ -1219,7 +1244,7 @@ std::string ARBDecompiler::VisitExpression(const Expr& node) {
         return Visit(ir.GetConditionCode(expr->cc));
     }
     if (const auto expr = std::get_if<ExprVar>(&*node)) {
-        return fmt::format("F{}.x", expr->var_index);
+        return fmt::format("F{}.x", expr->var_index + ast_var_base);
     }
     if (const auto expr = std::get_if<ExprBoolean>(&*node)) {
         return expr->value ? "0xffffffff" : "0";
@@ -1406,6 +1431,11 @@ std::string ARBDecompiler::Visit(const Node& node) {
         return {};
     }
 
+    if (const auto func_call = std::get_if<FunctionCallNode>(&*node)) {
+        AddLine("CAL func_{};", func_call->GetFuncId());
+        return {};
+    }
+
     if ([[maybe_unused]] const auto cmt = std::get_if<CommentNode>(&*node)) {
         // Uncommenting this will generate invalid code. GLASM lacks comments.
         // AddLine("// {}", cmt->GetText());
@@ -1479,7 +1509,7 @@ std::string ARBDecompiler::GlobalMemoryPointer(const GmemNode& gmem) {
 }
 
 void ARBDecompiler::Exit() {
-    if (stage != ShaderType::Fragment) {
+    if (!context_func->IsMain() || stage != ShaderType::Fragment) {
         AddLine("RET;");
         return;
     }
@@ -2021,13 +2051,13 @@ std::string ARBDecompiler::ImageStore(Operation operation) {
 
 std::string ARBDecompiler::Branch(Operation operation) {
     const auto target = std::get<ImmediateNode>(*operation[0]);
-    AddLine("MOV.U PC.x, {};", target.GetValue());
+    AddLine("MOV.U PC{}.x, {};", context_func->GetId(), target.GetValue());
     AddLine("CONT;");
     return {};
 }
 
 std::string ARBDecompiler::BranchIndirect(Operation operation) {
-    AddLine("MOV.U PC.x, {};", Visit(operation[0]));
+    AddLine("MOV.U PC{}.x, {};", context_func->GetId(), Visit(operation[0]));
     AddLine("CONT;");
     return {};
 }
@@ -2045,7 +2075,7 @@ std::string ARBDecompiler::PopFlowStack(Operation operation) {
     const auto stack = std::get<MetaStackClass>(operation.GetMeta());
     const std::string_view stack_name = StackName(stack);
     AddLine("SUB.S {}_TOP.x, {}_TOP.x, 1;", stack_name, stack_name);
-    AddLine("MOV.U PC.x, {}[{}_TOP.x].x;", stack_name, stack_name);
+    AddLine("MOV.U PC{}.x, {}[{}_TOP.x].x;", context_func->GetId(), stack_name, stack_name);
     AddLine("CONT;");
     return {};
 }
@@ -2056,6 +2086,10 @@ std::string ARBDecompiler::Exit(Operation) {
 }
 
 std::string ARBDecompiler::Discard(Operation) {
+    if (stage != ShaderType::Fragment) {
+        AddLine("RET;");
+        return {};
+    }
     AddLine("KIL TR;");
     return {};
 }

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -548,6 +548,26 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
     query_cache.FlushRegion(addr, size);
 }
 
+void RasterizerOpenGL::InvalidateExceptTextureCache(VAddr addr, u64 size) {
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    shader_cache.InvalidateRegion(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.WriteMemory(addr, size);
+    }
+    query_cache.InvalidateRegion(addr, size);
+}
+
+void RasterizerOpenGL::InvalidateTextureCache(VAddr addr, u64 size) {
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    std::scoped_lock lock{texture_cache.mutex};
+    texture_cache.UnmapMemory(addr, size);
+}
+
 bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size) {
     std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
     if (!Settings::IsGPULevelHigh()) {
@@ -611,6 +631,13 @@ void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) {
     shader_cache.OnCPUWrite(addr, size);
 }
 
+void RasterizerOpenGL::ModifyGPUMemory(GPUVAddr addr, u64 size) {
+    {
+        std::scoped_lock lock{texture_cache.mutex};
+        texture_cache.UnmapGPUMemory(addr, size);
+    }
+}
+
 void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) {
     if (!gpu.IsAsync()) {
         gpu_memory.Write<u32>(addr, value);

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -75,11 +75,14 @@ public:
     void DisableGraphicsUniformBuffer(size_t stage, u32 index) override;
     void FlushAll() override;
     void FlushRegion(VAddr addr, u64 size) override;
+    void InvalidateExceptTextureCache(VAddr addr, u64 size) override;
+    void InvalidateTextureCache(VAddr addr, u64 size) override;
     bool MustFlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;
     void OnCPUWrite(VAddr addr, u64 size) override;
     void SyncGuestHost() override;
     void UnmapMemory(VAddr addr, u64 size) override;
+    void ModifyGPUMemory(GPUVAddr addr, u64 size) override;
     void SignalSemaphore(GPUVAddr addr, u32 value) override;
     void SignalSyncPoint(u32 value) override;
     void ReleaseFences() override;

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -435,6 +435,27 @@ public:
         DeclareCustomVariables();
         DeclarePhysicalAttributeReader();
 
+        const auto& subfunctions = ir.GetSubFunctions();
+        auto it = subfunctions.rbegin();
+        while (it != subfunctions.rend()) {
+            context_func = *it;
+            code.AddLine("void func_{}() {{", context_func->GetId());
+            ++code.scope;
+
+            if (context_func->IsDecompiled()) {
+                DecompileAST();
+            } else {
+                DecompileBranchMode();
+            }
+
+            --code.scope;
+            code.AddLine("}}");
+
+            it++;
+        }
+
+        context_func = ir.GetMainFunction();
+
         code.AddLine("void main() {{");
         ++code.scope;
 
@@ -442,7 +463,7 @@ public:
             code.AddLine("gl_Position = vec4(0.0f, 0.0f, 0.0f, 1.0f);");
         }
 
-        if (ir.IsDecompiled()) {
+        if (context_func->IsDecompiled()) {
             DecompileAST();
         } else {
             DecompileBranchMode();
@@ -462,13 +483,13 @@ private:
 
     void DecompileBranchMode() {
         // VM's program counter
-        const auto first_address = ir.GetBasicBlocks().begin()->first;
+        const auto first_address = context_func->GetBasicBlocks().begin()->first;
         code.AddLine("uint jmp_to = {}U;", first_address);
 
         // TODO(Subv): Figure out the actual depth of the flow stack, for now it seems
         // unlikely that shaders will use 20 nested SSYs and PBKs.
         constexpr u32 FLOW_STACK_SIZE = 20;
-        if (!ir.IsFlowStackDisabled()) {
+        if (!context_func->IsFlowStackDisabled()) {
             for (const auto stack : std::array{MetaStackClass::Ssy, MetaStackClass::Pbk}) {
                 code.AddLine("uint {}[{}];", FlowStackName(stack), FLOW_STACK_SIZE);
                 code.AddLine("uint {} = 0U;", FlowStackTopName(stack));
@@ -480,7 +501,7 @@ private:
 
         code.AddLine("switch (jmp_to) {{");
 
-        for (const auto& pair : ir.GetBasicBlocks()) {
+        for (const auto& pair : context_func->GetBasicBlocks()) {
             const auto& [address, bb] = pair;
             code.AddLine("case 0x{:X}U: {{", address);
             ++code.scope;
@@ -1131,6 +1152,11 @@ private:
             return {};
         }
 
+        if (const auto func_call = std::get_if<FunctionCallNode>(&*node)) {
+            code.AddLine("func_{}();", func_call->GetFuncId());
+            return {};
+        }
+
         if (const auto comment = std::get_if<CommentNode>(&*node)) {
             code.AddLine("// " + comment->GetText());
             return {};
@@ -2267,7 +2293,9 @@ private:
     }
 
     Expression Exit(Operation operation) {
-        PreExit();
+        if (context_func->IsMain()) {
+            PreExit();
+        }
         code.AddLine("return;");
         return {};
     }
@@ -2277,7 +2305,11 @@ private:
         // about unexecuted instructions that may follow this.
         code.AddLine("if (true) {{");
         ++code.scope;
-        code.AddLine("discard;");
+        if (stage != ShaderType::Fragment) {
+            code.AddLine("return;");
+        } else {
+            code.AddLine("discard;");
+        }
         --code.scope;
         code.AddLine("}}");
         return {};
@@ -2388,7 +2420,7 @@ private:
     }
 
     Expression Barrier(Operation) {
-        if (!ir.IsDecompiled()) {
+        if (!context_func->IsDecompiled()) {
             LOG_ERROR(Render_OpenGL, "barrier() used but shader is not decompiled");
             return {};
         }
@@ -2755,6 +2787,8 @@ private:
     const Header header;
     std::unordered_map<u8, VaryingTFB> transform_feedback;
 
+    std::shared_ptr<ShaderFunctionIR> context_func;
+
     ShaderWriter code;
 
     std::optional<u32> max_input_vertices;
@@ -2902,9 +2936,15 @@ public:
             decomp.code.scope++;
         }
         if (ast.kills) {
-            decomp.code.AddLine("discard;");
+            if (decomp.stage != ShaderType::Fragment) {
+                decomp.code.AddLine("return;");
+            } else {
+                decomp.code.AddLine("discard;");
+            }
         } else {
-            decomp.PreExit();
+            if (decomp.context_func->IsMain()) {
+                decomp.PreExit();
+            }
             decomp.code.AddLine("return;");
         }
         if (!is_true) {
@@ -2937,13 +2977,13 @@ private:
 };
 
 void GLSLDecompiler::DecompileAST() {
-    const u32 num_flow_variables = ir.GetASTNumVariables();
+    const u32 num_flow_variables = context_func->GetASTNumVariables();
     for (u32 i = 0; i < num_flow_variables; i++) {
         code.AddLine("bool {} = false;", GetFlowVariable(i));
     }
 
     ASTDecompiler decompiler{*this};
-    decompiler.Visit(ir.GetASTProgram());
+    decompiler.Visit(context_func->GetASTProgram());
 }
 
 } // Anonymous namespace

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -327,7 +327,8 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4
     if (format_info.is_compressed) {
         return false;
     }
-    if (std::ranges::find(ACCELERATED_FORMATS, internal_format) == ACCELERATED_FORMATS.end()) {
+    if (std::ranges::find(ACCELERATED_FORMATS, static_cast<int>(internal_format)) ==
+        ACCELERATED_FORMATS.end()) {
         return false;
     }
     if (format_info.compatibility_by_size) {

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -497,6 +497,26 @@ void RasterizerVulkan::FlushRegion(VAddr addr, u64 size) {
     query_cache.FlushRegion(addr, size);
 }
 
+void Vulkan::RasterizerVulkan::InvalidateExceptTextureCache(VAddr addr, u64 size) {
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    pipeline_cache.InvalidateRegion(addr, size);
+    {
+        std::scoped_lock lock{buffer_cache.mutex};
+        buffer_cache.WriteMemory(addr, size);
+    }
+    query_cache.InvalidateRegion(addr, size);
+}
+
+void Vulkan::RasterizerVulkan::InvalidateTextureCache(VAddr addr, u64 size) {
+    if (addr == 0 || size == 0) {
+        return;
+    }
+    std::scoped_lock lock{texture_cache.mutex};
+    texture_cache.UnmapMemory(addr, size);
+}
+
 bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size) {
     std::scoped_lock lock{texture_cache.mutex, buffer_cache.mutex};
     if (!Settings::IsGPULevelHigh()) {
@@ -557,6 +577,13 @@ void RasterizerVulkan::UnmapMemory(VAddr addr, u64 size) {
     pipeline_cache.OnCPUWrite(addr, size);
 }
 
+void RasterizerVulkan::ModifyGPUMemory(GPUVAddr addr, u64 size) {
+    {
+        std::scoped_lock lock{texture_cache.mutex};
+        texture_cache.UnmapGPUMemory(addr, size);
+    }
+}
+
 void RasterizerVulkan::SignalSemaphore(GPUVAddr addr, u32 value) {
     if (!gpu.IsAsync()) {
         gpu_memory.Write<u32>(addr, value);

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -67,11 +67,14 @@ public:
     void DisableGraphicsUniformBuffer(size_t stage, u32 index) override;
     void FlushAll() override;
     void FlushRegion(VAddr addr, u64 size) override;
+    void InvalidateExceptTextureCache(VAddr addr, u64 size) override;
+    void InvalidateTextureCache(VAddr addr, u64 size) override;
     bool MustFlushRegion(VAddr addr, u64 size) override;
     void InvalidateRegion(VAddr addr, u64 size) override;
     void OnCPUWrite(VAddr addr, u64 size) override;
     void SyncGuestHost() override;
     void UnmapMemory(VAddr addr, u64 size) override;
+    void ModifyGPUMemory(GPUVAddr addr, u64 size) override;
     void SignalSemaphore(GPUVAddr addr, u32 value) override;
     void SignalSyncPoint(u32 value) override;
     void ReleaseFences() override;

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -406,10 +406,38 @@ private:
         binding = DeclareStorageTexels(binding);
         binding = DeclareImages(binding);
 
+        const auto& subfunctions = ir.GetSubFunctions();
+
+        labels.resize(subfunctions.size() + 1);
+        other_functions.resize(subfunctions.size());
+
+        auto it = subfunctions.rbegin();
+        while (it != subfunctions.rend()) {
+            context_func = *it;
+            other_functions[context_func->GetId() - 1] =
+                OpFunction(t_void, {}, TypeFunction(t_void));
+            AddLabel();
+
+            if (context_func->IsDecompiled()) {
+                DeclareFlowVariables();
+                DecompileAST();
+            } else {
+                AllocateLabels();
+                DecompileBranchMode();
+            }
+
+            OpReturn();
+            OpFunctionEnd();
+
+            it++;
+        }
+
+        context_func = ir.GetMainFunction();
+
         const Id main = OpFunction(t_void, {}, TypeFunction(t_void));
         AddLabel();
 
-        if (ir.IsDecompiled()) {
+        if (context_func->IsDecompiled()) {
             DeclareFlowVariables();
             DecompileAST();
         } else {
@@ -441,16 +469,18 @@ private:
     void DecompileAST();
 
     void DecompileBranchMode() {
-        const u32 first_address = ir.GetBasicBlocks().begin()->first;
-        const Id loop_label = OpLabel("loop");
-        const Id merge_label = OpLabel("merge");
+        const u32 first_address = context_func->GetBasicBlocks().begin()->first;
+        const u32 func_id = context_func->GetId();
+        const std::string func_id_msg = std::to_string(func_id);
+        const Id loop_label = OpLabel("loop_" + func_id_msg);
+        const Id merge_label = OpLabel("merge_" + func_id_msg);
         const Id dummy_label = OpLabel();
         const Id jump_label = OpLabel();
-        continue_label = OpLabel("continue");
+        continue_label = OpLabel("continue_" + func_id_msg);
 
         std::vector<Sirit::Literal> literals;
         std::vector<Id> branch_labels;
-        for (const auto& [literal, label] : labels) {
+        for (const auto& [literal, label] : labels[func_id]) {
             literals.push_back(literal);
             branch_labels.push_back(label);
         }
@@ -462,11 +492,11 @@ private:
         std::tie(ssy_flow_stack, ssy_flow_stack_top) = CreateFlowStack();
         std::tie(pbk_flow_stack, pbk_flow_stack_top) = CreateFlowStack();
 
-        Name(jmp_to, "jmp_to");
-        Name(ssy_flow_stack, "ssy_flow_stack");
-        Name(ssy_flow_stack_top, "ssy_flow_stack_top");
-        Name(pbk_flow_stack, "pbk_flow_stack");
-        Name(pbk_flow_stack_top, "pbk_flow_stack_top");
+        Name(jmp_to, "jmp_to_" + func_id_msg);
+        Name(ssy_flow_stack, "ssy_flow_stack_" + func_id_msg);
+        Name(ssy_flow_stack_top, "ssy_flow_stack_top_" + func_id_msg);
+        Name(pbk_flow_stack, "pbk_flow_stack_" + func_id_msg);
+        Name(pbk_flow_stack_top, "pbk_flow_stack_top_" + func_id_msg);
 
         DefinePrologue();
 
@@ -484,13 +514,14 @@ private:
         AddLabel(default_branch);
         OpReturn();
 
-        for (const auto& [address, bb] : ir.GetBasicBlocks()) {
-            AddLabel(labels.at(address));
+        for (const auto& [address, bb] : context_func->GetBasicBlocks()) {
+            AddLabel(labels[func_id].at(address));
 
             VisitBasicBlock(bb);
 
-            const auto next_it = labels.lower_bound(address + 1);
-            const Id next_label = next_it != labels.end() ? next_it->second : default_branch;
+            const auto next_it = labels[func_id].lower_bound(address + 1);
+            const Id next_label =
+                next_it != labels[func_id].end() ? next_it->second : default_branch;
             OpBranch(next_label);
         }
 
@@ -508,9 +539,10 @@ private:
     static constexpr auto INTERNAL_FLAGS_COUNT = static_cast<std::size_t>(InternalFlag::Amount);
 
     void AllocateLabels() {
-        for (const auto& pair : ir.GetBasicBlocks()) {
+        const u32 func_id = context_func->GetId();
+        for (const auto& pair : context_func->GetBasicBlocks()) {
             const u32 address = pair.first;
-            labels.emplace(address, OpLabel(fmt::format("label_0x{:x}", address)));
+            labels[func_id].emplace(address, OpLabel(fmt::format("label_0x{:x}", address)));
         }
     }
 
@@ -589,6 +621,14 @@ private:
         DeclareOutputVertex();
     }
 
+    void SafeKill() {
+        if (stage != ShaderType::Fragment) {
+            OpReturn();
+            return;
+        }
+        OpKill();
+    }
+
     void DeclareFragment() {
         if (stage != ShaderType::Fragment) {
             return;
@@ -656,7 +696,7 @@ private:
     }
 
     void DeclareFlowVariables() {
-        for (u32 i = 0; i < ir.GetASTNumVariables(); i++) {
+        for (u32 i = 0; i < context_func->GetASTNumVariables(); i++) {
             const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false);
             Name(id, fmt::format("flow_var_{}", static_cast<u32>(i)));
             flow_variables.emplace(i, AddGlobalVariable(id));
@@ -1333,6 +1373,12 @@ private:
             return {};
         }
 
+        if (const auto func_call = std::get_if<FunctionCallNode>(&*node)) {
+            const u32 func_id = func_call->GetFuncId();
+            OpFunctionCall(t_void, other_functions[func_id - 1]);
+            return {};
+        }
+
         if (const auto comment = std::get_if<CommentNode>(&*node)) {
             if (device.HasDebuggingToolAttached()) {
                 // We should insert comments with OpString instead of using named variables
@@ -2124,7 +2170,7 @@ private:
 
         OpBranchConditional(condition, true_label, discard_label);
         AddLabel(discard_label);
-        OpKill();
+        SafeKill();
         AddLabel(true_label);
     }
 
@@ -2175,7 +2221,9 @@ private:
     }
 
     Expression Exit(Operation operation) {
-        PreExit();
+        if (context_func->IsMain()) {
+            PreExit();
+        }
         inside_branch = true;
         if (conditional_branch_set) {
             OpReturn();
@@ -2192,12 +2240,12 @@ private:
     Expression Discard(Operation operation) {
         inside_branch = true;
         if (conditional_branch_set) {
-            OpKill();
+            SafeKill();
         } else {
             const Id dummy = OpLabel();
             OpBranch(dummy);
             AddLabel(dummy);
-            OpKill();
+            SafeKill();
             AddLabel();
         }
         return {};
@@ -2276,7 +2324,7 @@ private:
     }
 
     Expression Barrier(Operation) {
-        if (!ir.IsDecompiled()) {
+        if (!context_func->IsDecompiled()) {
             LOG_ERROR(Render_Vulkan, "OpBarrier used by shader is not decompiled");
             return {};
         }
@@ -2770,6 +2818,8 @@ private:
     const Specialization& specialization;
     std::unordered_map<u8, VaryingTFB> transform_feedback;
 
+    std::shared_ptr<ShaderFunctionIR> context_func;
+
     const Id t_void = Name(TypeVoid(), "void");
 
     const Id t_bool = Name(TypeBool(), "bool");
@@ -2896,7 +2946,8 @@ private:
     Id ssy_flow_stack{};
     Id pbk_flow_stack{};
     Id continue_label{};
-    std::map<u32, Id> labels;
+    std::vector<std::map<u32, Id>> labels;
+    std::vector<Id> other_functions;
 
     bool conditional_branch_set{};
     bool inside_branch{};
@@ -3047,9 +3098,11 @@ public:
             decomp.OpBranchConditional(condition, then_label, endif_label);
             decomp.AddLabel(then_label);
             if (ast.kills) {
-                decomp.OpKill();
+                decomp.SafeKill();
             } else {
-                decomp.PreExit();
+                if (decomp.context_func->IsMain()) {
+                    decomp.PreExit();
+                }
                 decomp.OpReturn();
             }
             decomp.AddLabel(endif_label);
@@ -3058,9 +3111,11 @@ public:
             decomp.OpBranch(next_block);
             decomp.AddLabel(next_block);
             if (ast.kills) {
-                decomp.OpKill();
+                decomp.SafeKill();
             } else {
-                decomp.PreExit();
+                if (decomp.context_func->IsMain()) {
+                    decomp.PreExit();
+                }
                 decomp.OpReturn();
             }
             decomp.AddLabel(decomp.OpLabel());
@@ -3097,7 +3152,7 @@ private:
 };
 
 void SPIRVDecompiler::DecompileAST() {
-    const u32 num_flow_variables = ir.GetASTNumVariables();
+    const u32 num_flow_variables = context_func->GetASTNumVariables();
     for (u32 i = 0; i < num_flow_variables; i++) {
         const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false);
         Name(id, fmt::format("flow_var_{}", i));
@@ -3106,7 +3161,7 @@ void SPIRVDecompiler::DecompileAST() {
 
     DefinePrologue();
 
-    const ASTNode program = ir.GetASTProgram();
+    const ASTNode program = context_func->GetASTProgram();
     ASTDecompiler decompiler{*this};
     decompiler.Visit(program);
 

src/video_core/shader/control_flow.cpp

@@ -7,6 +7,7 @@
 #include <set>
 #include <stack>
 #include <unordered_map>
+#include <unordered_set>
 #include <vector>
 
 #include "common/assert.h"
@@ -26,17 +27,29 @@ using Tegra::Shader::OpCode;
 
 constexpr s32 unassigned_branch = -2;
 
+enum class JumpLabel : u32 {
+    SSYClass = 0,
+    PBKClass = 1,
+};
+
+struct JumpItem {
+    JumpLabel type;
+    u32 address;
+
+    bool operator==(const JumpItem& other) const {
+        return std::tie(type, address) == std::tie(other.type, other.address);
+    }
+};
+
 struct Query {
     u32 address{};
-    std::stack<u32> ssy_stack{};
-    std::stack<u32> pbk_stack{};
+    std::stack<JumpItem> stack{};
 };
 
 struct BlockStack {
     BlockStack() = default;
-    explicit BlockStack(const Query& q) : ssy_stack{q.ssy_stack}, pbk_stack{q.pbk_stack} {}
-    std::stack<u32> ssy_stack{};
-    std::stack<u32> pbk_stack{};
+    explicit BlockStack(const Query& q) : stack{q.stack} {}
+    std::stack<JumpItem> stack{};
 };
 
 template <typename T, typename... Args>
@@ -65,20 +78,36 @@ struct BlockInfo {
     }
 };
 
-struct CFGRebuildState {
-    explicit CFGRebuildState(const ProgramCode& program_code_, u32 start_, Registry& registry_)
-        : program_code{program_code_}, registry{registry_}, start{start_} {}
+struct ProgramControl {
+    std::unordered_set<u32> found_functions{};
+    std::list<u32> pending_functions{};
 
+    void RegisterFunction(u32 address) {
+        if (found_functions.count(address) != 0) {
+            return;
+        }
+        found_functions.insert(address);
+        pending_functions.emplace_back(address);
+    }
+};
+
+struct CFGRebuildState {
+    explicit CFGRebuildState(ProgramControl& control_, const ProgramCode& program_code_, u32 start_,
+                             u32 base_start_, Registry& registry_)
+        : control{control_}, program_code{program_code_}, registry{registry_}, start{start_},
+          base_start{base_start_} {}
+
+    ProgramControl& control;
     const ProgramCode& program_code;
     Registry& registry;
     u32 start{};
+    u32 base_start{};
     std::vector<BlockInfo> block_info;
     std::list<u32> inspect_queries;
     std::list<Query> queries;
     std::unordered_map<u32, u32> registered;
     std::set<u32> labels;
-    std::map<u32, u32> ssy_labels;
-    std::map<u32, u32> pbk_labels;
+    std::map<u32, JumpItem> jump_labels;
     std::unordered_map<u32, BlockStack> stacks;
     ASTManager* manager{};
 };
@@ -153,7 +182,7 @@ template <typename Result, typename TestCallable, typename PackCallable>
 std::optional<Result> TrackInstruction(const CFGRebuildState& state, u32& pos, TestCallable test,
                                        PackCallable pack) {
     for (; pos >= state.start; --pos) {
-        if (IsSchedInstruction(pos, state.start)) {
+        if (IsSchedInstruction(pos, state.base_start)) {
             continue;
         }
         const Instruction instr = state.program_code[pos];
@@ -262,7 +291,7 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
             single_branch.ignore = true;
             break;
         }
-        if (IsSchedInstruction(offset, state.start)) {
+        if (IsSchedInstruction(offset, state.base_start)) {
             offset++;
             continue;
         }
@@ -274,6 +303,7 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
         }
 
         switch (opcode->get().GetId()) {
+        case OpCode::Id::RET:
         case OpCode::Id::EXIT: {
             const auto pred_index = static_cast<u32>(instr.pred.pred_index);
             single_branch.condition.predicate = GetPredicate(pred_index, instr.negate_pred != 0);
@@ -411,13 +441,20 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
         case OpCode::Id::SSY: {
             const u32 target = offset + instr.bra.GetBranchTarget();
             insert_label(state, target);
-            state.ssy_labels.emplace(offset, target);
+            JumpItem it = {JumpLabel::SSYClass, target};
+            state.jump_labels.emplace(offset, it);
             break;
         }
         case OpCode::Id::PBK: {
             const u32 target = offset + instr.bra.GetBranchTarget();
             insert_label(state, target);
-            state.pbk_labels.emplace(offset, target);
+            JumpItem it = {JumpLabel::PBKClass, target};
+            state.jump_labels.emplace(offset, it);
+            break;
+        }
+        case OpCode::Id::CAL: {
+            const u32 target = offset + instr.bra.GetBranchTarget();
+            state.control.RegisterFunction(target);
             break;
         }
         case OpCode::Id::BRX: {
@@ -513,7 +550,7 @@ bool TryInspectAddress(CFGRebuildState& state) {
 }
 
 bool TryQuery(CFGRebuildState& state) {
-    const auto gather_labels = [](std::stack<u32>& cc, std::map<u32, u32>& labels,
+    const auto gather_labels = [](std::stack<JumpItem>& cc, std::map<u32, JumpItem>& labels,
                                   BlockInfo& block) {
         auto gather_start = labels.lower_bound(block.start);
         const auto gather_end = labels.upper_bound(block.end);
@@ -522,6 +559,19 @@ bool TryQuery(CFGRebuildState& state) {
             ++gather_start;
         }
     };
+    const auto pop_labels = [](JumpLabel type, SingleBranch* branch, Query& query) -> bool {
+        while (!query.stack.empty() && query.stack.top().type != type) {
+            query.stack.pop();
+        }
+        if (query.stack.empty()) {
+            return false;
+        }
+        if (branch->address == unassigned_branch) {
+            branch->address = query.stack.top().address;
+        }
+        query.stack.pop();
+        return true;
+    };
     if (state.queries.empty()) {
         return false;
     }
@@ -534,8 +584,7 @@ bool TryQuery(CFGRebuildState& state) {
     // consumes a label. Schedule new queries accordingly
     if (block.visited) {
         BlockStack& stack = state.stacks[q.address];
-        const bool all_okay = (stack.ssy_stack.empty() || q.ssy_stack == stack.ssy_stack) &&
-                              (stack.pbk_stack.empty() || q.pbk_stack == stack.pbk_stack);
+        const bool all_okay = (stack.stack.empty() || q.stack == stack.stack);
         state.queries.pop_front();
         return all_okay;
     }
@@ -544,8 +593,7 @@ bool TryQuery(CFGRebuildState& state) {
 
     Query q2(q);
     state.queries.pop_front();
-    gather_labels(q2.ssy_stack, state.ssy_labels, block);
-    gather_labels(q2.pbk_stack, state.pbk_labels, block);
+    gather_labels(q2.stack, state.jump_labels, block);
     if (std::holds_alternative<SingleBranch>(*block.branch)) {
         auto* branch = std::get_if<SingleBranch>(block.branch.get());
         if (!branch->condition.IsUnconditional()) {
@@ -555,16 +603,10 @@ bool TryQuery(CFGRebuildState& state) {
 
         auto& conditional_query = state.queries.emplace_back(q2);
         if (branch->is_sync) {
-            if (branch->address == unassigned_branch) {
-                branch->address = conditional_query.ssy_stack.top();
-            }
-            conditional_query.ssy_stack.pop();
+            pop_labels(JumpLabel::SSYClass, branch, conditional_query);
         }
         if (branch->is_brk) {
-            if (branch->address == unassigned_branch) {
-                branch->address = conditional_query.pbk_stack.top();
-            }
-            conditional_query.pbk_stack.pop();
+            pop_labels(JumpLabel::PBKClass, branch, conditional_query);
         }
         conditional_query.address = branch->address;
         return true;
@@ -646,25 +688,23 @@ void DecompileShader(CFGRebuildState& state) {
     state.manager->Decompile();
 }
 
-} // Anonymous namespace
-
-std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
-                                                const CompilerSettings& settings,
-                                                Registry& registry) {
-    auto result_out = std::make_unique<ShaderCharacteristics>();
+ShaderFunction ScanFunction(ProgramControl& control, const ProgramCode& program_code,
+                            u32 start_address, u32 base_start, const CompilerSettings& settings,
+                            Registry& registry) {
+    ShaderFunction result_out{};
     if (settings.depth == CompileDepth::BruteForce) {
-        result_out->settings.depth = CompileDepth::BruteForce;
+        result_out.settings.depth = CompileDepth::BruteForce;
         return result_out;
     }
 
-    CFGRebuildState state{program_code, start_address, registry};
+    CFGRebuildState state{control, program_code, start_address, base_start, registry};
     // Inspect Code and generate blocks
     state.labels.clear();
     state.labels.emplace(start_address);
     state.inspect_queries.push_back(state.start);
     while (!state.inspect_queries.empty()) {
         if (!TryInspectAddress(state)) {
-            result_out->settings.depth = CompileDepth::BruteForce;
+            result_out.settings.depth = CompileDepth::BruteForce;
             return result_out;
         }
     }
@@ -675,7 +715,7 @@ std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
 
     if (settings.depth != CompileDepth::FlowStack) {
         // Decompile Stacks
-        state.queries.push_back(Query{state.start, {}, {}});
+        state.queries.push_back(Query{state.start, {}});
         decompiled = true;
         while (!state.queries.empty()) {
             if (!TryQuery(state)) {
@@ -705,19 +745,18 @@ std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
             state.manager->ShowCurrentState("Of Shader");
             state.manager->Clear();
         } else {
-            auto characteristics = std::make_unique<ShaderCharacteristics>();
-            characteristics->start = start_address;
-            characteristics->settings.depth = settings.depth;
-            characteristics->manager = std::move(manager);
-            characteristics->end = state.block_info.back().end + 1;
-            return characteristics;
+            result_out.start = start_address;
+            result_out.settings.depth = settings.depth;
+            result_out.manager = std::move(manager);
+            result_out.end = state.block_info.back().end + 1;
+            return result_out;
         }
     }
 
-    result_out->start = start_address;
-    result_out->settings.depth =
+    result_out.start = start_address;
+    result_out.settings.depth =
         use_flow_stack ? CompileDepth::FlowStack : CompileDepth::NoFlowStack;
-    result_out->blocks.clear();
+    result_out.blocks.clear();
     for (auto& block : state.block_info) {
         ShaderBlock new_block{};
         new_block.start = block.start;
@@ -726,20 +765,20 @@ std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
         if (!new_block.ignore_branch) {
             new_block.branch = block.branch;
         }
-        result_out->end = std::max(result_out->end, block.end);
-        result_out->blocks.push_back(new_block);
+        result_out.end = std::max(result_out.end, block.end);
+        result_out.blocks.push_back(new_block);
     }
     if (!use_flow_stack) {
-        result_out->labels = std::move(state.labels);
+        result_out.labels = std::move(state.labels);
         return result_out;
     }
 
-    auto back = result_out->blocks.begin();
+    auto back = result_out.blocks.begin();
     auto next = std::next(back);
-    while (next != result_out->blocks.end()) {
+    while (next != result_out.blocks.end()) {
         if (!state.labels.contains(next->start) && next->start == back->end + 1) {
             back->end = next->end;
-            next = result_out->blocks.erase(next);
+            next = result_out.blocks.erase(next);
             continue;
         }
         back = next;
@@ -748,4 +787,22 @@ std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code,
 
     return result_out;
 }
+
+} // Anonymous namespace
+
+std::unique_ptr<ShaderProgram> ScanFlow(const ProgramCode& program_code, u32 start_address,
+                                        const CompilerSettings& settings, Registry& registry) {
+    ProgramControl control{};
+    auto result_out = std::make_unique<ShaderProgram>();
+    result_out->main =
+        ScanFunction(control, program_code, start_address, start_address, settings, registry);
+    while (!control.pending_functions.empty()) {
+        u32 address = control.pending_functions.front();
+        auto fun = ScanFunction(control, program_code, address, start_address, settings, registry);
+        result_out->subfunctions.emplace(address, std::move(fun));
+        control.pending_functions.pop_front();
+    }
+    return result_out;
+}
+
 } // namespace VideoCommon::Shader

src/video_core/shader/control_flow.h

@@ -5,6 +5,7 @@
 #pragma once
 
 #include <list>
+#include <map>
 #include <optional>
 #include <set>
 #include <variant>
@@ -101,7 +102,7 @@ struct ShaderBlock {
     }
 };
 
-struct ShaderCharacteristics {
+struct ShaderFunction {
     std::list<ShaderBlock> blocks{};
     std::set<u32> labels{};
     u32 start{};
@@ -110,8 +111,12 @@ struct ShaderCharacteristics {
     CompilerSettings settings{};
 };
 
-std::unique_ptr<ShaderCharacteristics> ScanFlow(const ProgramCode& program_code, u32 start_address,
-                                                const CompilerSettings& settings,
-                                                Registry& registry);
+struct ShaderProgram {
+    ShaderFunction main;
+    std::map<u32, ShaderFunction> subfunctions;
+};
+
+std::unique_ptr<ShaderProgram> ScanFlow(const ProgramCode& program_code, u32 start_address,
+                                        const CompilerSettings& settings, Registry& registry);
 
 } // namespace VideoCommon::Shader

src/video_core/shader/decode.cpp

@@ -64,9 +64,52 @@ std::optional<u32> TryDeduceSamplerSize(const SamplerEntry& sampler_to_deduce,
 
 } // Anonymous namespace
 
+class ExprDecoder {
+public:
+    explicit ExprDecoder(ShaderIR& ir_) : ir(ir_) {}
+
+    void operator()(const ExprAnd& expr) {
+        Visit(expr.operand1);
+        Visit(expr.operand2);
+    }
+
+    void operator()(const ExprOr& expr) {
+        Visit(expr.operand1);
+        Visit(expr.operand2);
+    }
+
+    void operator()(const ExprNot& expr) {
+        Visit(expr.operand1);
+    }
+
+    void operator()(const ExprPredicate& expr) {
+        const auto pred = static_cast<Tegra::Shader::Pred>(expr.predicate);
+        if (pred != Pred::UnusedIndex && pred != Pred::NeverExecute) {
+            ir.used_predicates.insert(pred);
+        }
+    }
+
+    void operator()(const ExprCondCode& expr) {}
+
+    void operator()(const ExprVar& expr) {}
+
+    void operator()(const ExprBoolean& expr) {}
+
+    void operator()(const ExprGprEqual& expr) {
+        ir.used_registers.insert(expr.gpr);
+    }
+
+    void Visit(const Expr& node) {
+        return std::visit(*this, *node);
+    }
+
+private:
+    ShaderIR& ir;
+};
+
 class ASTDecoder {
 public:
-    explicit ASTDecoder(ShaderIR& ir_) : ir(ir_) {}
+    explicit ASTDecoder(ShaderIR& ir_) : ir(ir_), decoder(ir_) {}
 
     void operator()(ASTProgram& ast) {
         ASTNode current = ast.nodes.GetFirst();
@@ -77,6 +120,7 @@ public:
     }
 
     void operator()(ASTIfThen& ast) {
+        decoder.Visit(ast.condition);
         ASTNode current = ast.nodes.GetFirst();
         while (current) {
             Visit(current);
@@ -96,13 +140,18 @@ public:
 
     void operator()(ASTBlockDecoded& ast) {}
 
-    void operator()(ASTVarSet& ast) {}
+    void operator()(ASTVarSet& ast) {
+        decoder.Visit(ast.condition);
+    }
 
     void operator()(ASTLabel& ast) {}
 
-    void operator()(ASTGoto& ast) {}
+    void operator()(ASTGoto& ast) {
+        decoder.Visit(ast.condition);
+    }
 
     void operator()(ASTDoWhile& ast) {
+        decoder.Visit(ast.condition);
         ASTNode current = ast.nodes.GetFirst();
         while (current) {
             Visit(current);
@@ -110,9 +159,13 @@ public:
         }
     }
 
-    void operator()(ASTReturn& ast) {}
+    void operator()(ASTReturn& ast) {
+        decoder.Visit(ast.condition);
+    }
 
-    void operator()(ASTBreak& ast) {}
+    void operator()(ASTBreak& ast) {
+        decoder.Visit(ast.condition);
+    }
 
     void Visit(ASTNode& node) {
         std::visit(*this, *node->GetInnerData());
@@ -125,77 +178,113 @@ public:
 
 private:
     ShaderIR& ir;
+    ExprDecoder decoder;
 };
 
 void ShaderIR::Decode() {
+    const auto decode_function = ([this](ShaderFunction& shader_info) {
+        coverage_end = std::max<u32>(0, shader_info.end);
+        switch (shader_info.settings.depth) {
+        case CompileDepth::FlowStack: {
+            for (const auto& block : shader_info.blocks) {
+                basic_blocks.insert({block.start, DecodeRange(block.start, block.end + 1)});
+            }
+            break;
+        }
+        case CompileDepth::NoFlowStack: {
+            disable_flow_stack = true;
+            const auto insert_block = [this](NodeBlock& nodes, u32 label) {
+                if (label == static_cast<u32>(exit_branch)) {
+                    return;
+                }
+                basic_blocks.insert({label, nodes});
+            };
+            const auto& blocks = shader_info.blocks;
+            NodeBlock current_block;
+            u32 current_label = static_cast<u32>(exit_branch);
+            for (const auto& block : blocks) {
+                if (shader_info.labels.contains(block.start)) {
+                    insert_block(current_block, current_label);
+                    current_block.clear();
+                    current_label = block.start;
+                }
+                if (!block.ignore_branch) {
+                    DecodeRangeInner(current_block, block.start, block.end);
+                    InsertControlFlow(current_block, block);
+                } else {
+                    DecodeRangeInner(current_block, block.start, block.end + 1);
+                }
+            }
+            insert_block(current_block, current_label);
+            break;
+        }
+        case CompileDepth::DecompileBackwards:
+        case CompileDepth::FullDecompile: {
+            program_manager = std::move(shader_info.manager);
+            disable_flow_stack = true;
+            decompiled = true;
+            ASTDecoder decoder{*this};
+            ASTNode program = program_manager.GetProgram();
+            decoder.Visit(program);
+            break;
+        }
+        default:
+            LOG_CRITICAL(HW_GPU, "Unknown decompilation mode!");
+            [[fallthrough]];
+        case CompileDepth::BruteForce: {
+            const auto shader_end = static_cast<u32>(program_code.size());
+            coverage_begin = main_offset;
+            coverage_end = shader_end;
+            for (u32 label = main_offset; label < shader_end; ++label) {
+                basic_blocks.insert({label, DecodeRange(label, label + 1)});
+            }
+            break;
+        }
+        }
+        if (settings.depth != shader_info.settings.depth) {
+            LOG_WARNING(
+                HW_GPU,
+                "Decompiling to this setting \"{}\" failed, downgrading to this setting \"{}\"",
+                CompileDepthAsString(settings.depth),
+                CompileDepthAsString(shader_info.settings.depth));
+        }
+    });
+    const auto gen_function =
+        ([this](ShaderFunction& shader_info, u32 id) -> std::shared_ptr<ShaderFunctionIR> {
+            std::shared_ptr<ShaderFunctionIR> result;
+            if (decompiled) {
+                result = std::make_shared<ShaderFunctionIR>(std::move(program_manager), id,
+                                                            shader_info.start, shader_info.end);
+            } else {
+                result =
+                    std::make_shared<ShaderFunctionIR>(std::move(basic_blocks), disable_flow_stack,
+                                                       id, shader_info.start, shader_info.end);
+            }
+            decompiled = false;
+            disable_flow_stack = false;
+            basic_blocks.clear();
+            program_manager.Clear();
+            return result;
+        });
     std::memcpy(&header, program_code.data(), sizeof(Tegra::Shader::Header));
 
     decompiled = false;
     auto info = ScanFlow(program_code, main_offset, settings, registry);
-    auto& shader_info = *info;
-    coverage_begin = shader_info.start;
-    coverage_end = shader_info.end;
-    switch (shader_info.settings.depth) {
-    case CompileDepth::FlowStack: {
-        for (const auto& block : shader_info.blocks) {
-            basic_blocks.insert({block.start, DecodeRange(block.start, block.end + 1)});
-        }
-        break;
+    u32 id_start = 1;
+    for (auto& pair : info->subfunctions) {
+        func_map.emplace(pair.first, id_start);
+        id_start++;
     }
-    case CompileDepth::NoFlowStack: {
-        disable_flow_stack = true;
-        const auto insert_block = [this](NodeBlock& nodes, u32 label) {
-            if (label == static_cast<u32>(exit_branch)) {
-                return;
-            }
-            basic_blocks.insert({label, nodes});
-        };
-        const auto& blocks = shader_info.blocks;
-        NodeBlock current_block;
-        u32 current_label = static_cast<u32>(exit_branch);
-        for (const auto& block : blocks) {
-            if (shader_info.labels.contains(block.start)) {
-                insert_block(current_block, current_label);
-                current_block.clear();
-                current_label = block.start;
-            }
-            if (!block.ignore_branch) {
-                DecodeRangeInner(current_block, block.start, block.end);
-                InsertControlFlow(current_block, block);
-            } else {
-                DecodeRangeInner(current_block, block.start, block.end + 1);
-            }
-        }
-        insert_block(current_block, current_label);
-        break;
-    }
-    case CompileDepth::DecompileBackwards:
-    case CompileDepth::FullDecompile: {
-        program_manager = std::move(shader_info.manager);
-        disable_flow_stack = true;
-        decompiled = true;
-        ASTDecoder decoder{*this};
-        ASTNode program = GetASTProgram();
-        decoder.Visit(program);
-        break;
-    }
-    default:
-        LOG_CRITICAL(HW_GPU, "Unknown decompilation mode!");
-        [[fallthrough]];
-    case CompileDepth::BruteForce: {
-        const auto shader_end = static_cast<u32>(program_code.size());
-        coverage_begin = main_offset;
-        coverage_end = shader_end;
-        for (u32 label = main_offset; label < shader_end; ++label) {
-            basic_blocks.insert({label, DecodeRange(label, label + 1)});
-        }
-        break;
-    }
-    }
-    if (settings.depth != shader_info.settings.depth) {
-        LOG_WARNING(
-            HW_GPU, "Decompiling to this setting \"{}\" failed, downgrading to this setting \"{}\"",
-            CompileDepthAsString(settings.depth), CompileDepthAsString(shader_info.settings.depth));
+    coverage_begin = info->main.start;
+    coverage_end = 0;
+    decode_function(info->main);
+    main_function = gen_function(info->main, 0);
+    subfunctions.resize(info->subfunctions.size());
+    for (auto& pair : info->subfunctions) {
+        auto& func_info = pair.second;
+        decode_function(func_info);
+        u32 id = func_map[pair.first];
+        subfunctions[id - 1] = gen_function(func_info, id);
     }
 }
 

src/video_core/shader/decode/other.cpp

@@ -33,6 +33,7 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         // With the previous preconditions, this instruction is a no-operation.
         break;
     }
+    case OpCode::Id::RET:
     case OpCode::Id::EXIT: {
         const ConditionCode cc = instr.flow_condition_code;
         UNIMPLEMENTED_IF_MSG(cc != ConditionCode::T, "EXIT condition code used: {}", cc);
@@ -312,6 +313,16 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
         LOG_DEBUG(HW_GPU, "DEPBAR instruction is stubbed");
         break;
     }
+    case OpCode::Id::CAL: {
+        const u32 target = pc + instr.bra.GetBranchTarget();
+        const auto it = func_map.find(target);
+        if (it == func_map.end()) {
+            UNREACHABLE();
+            break;
+        }
+        bb.push_back(FunctionCall(it->second));
+        break;
+    }
     default:
         UNIMPLEMENTED_MSG("Unhandled instruction: {}", opcode->get().GetName());
     }

src/video_core/shader/node.h

@@ -267,10 +267,11 @@ class PatchNode;
 class SmemNode;
 class GmemNode;
 class CommentNode;
+class FunctionCallNode;
 
 using NodeData = std::variant<OperationNode, ConditionalNode, GprNode, CustomVarNode, ImmediateNode,
                               InternalFlagNode, PredicateNode, AbufNode, PatchNode, CbufNode,
-                              LmemNode, SmemNode, GmemNode, CommentNode>;
+                              LmemNode, SmemNode, GmemNode, FunctionCallNode, CommentNode>;
 using Node = std::shared_ptr<NodeData>;
 using Node4 = std::array<Node, 4>;
 using NodeBlock = std::vector<Node>;
@@ -494,6 +495,18 @@ private:
     std::vector<Node> code; ///< Code to execute
 };
 
+class FunctionCallNode final : public AmendNode {
+public:
+    explicit FunctionCallNode(u32 func_id_) : func_id{func_id_} {}
+
+    [[nodiscard]] u32 GetFuncId() const {
+        return func_id;
+    }
+
+private:
+    u32 func_id; ///< Id of the function to call
+};
+
 /// A general purpose register
 class GprNode final {
 public:

src/video_core/shader/node_helper.cpp

@@ -19,6 +19,11 @@ Node Comment(std::string text) {
     return MakeNode<CommentNode>(std::move(text));
 }
 
+/// Creates a function call
+Node FunctionCall(u32 func_id) {
+    return MakeNode<FunctionCallNode>(func_id);
+}
+
 Node Immediate(u32 value) {
     return MakeNode<ImmediateNode>(value);
 }

src/video_core/shader/node_helper.h

@@ -27,6 +27,9 @@ Node Conditional(Node condition, std::vector<Node> code);
 /// Creates a commentary node
 Node Comment(std::string text);
 
+/// Creates a function call
+Node FunctionCall(u32 func_id);
+
 /// Creates an u32 immediate
 Node Immediate(u32 value);
 

src/video_core/shader/shader_ir.h

@@ -26,7 +26,7 @@ namespace VideoCommon::Shader {
 
 struct ShaderBlock;
 
-constexpr u32 MAX_PROGRAM_LENGTH = 0x1000;
+constexpr u32 MAX_PROGRAM_LENGTH = 0x2000;
 
 struct ConstBuffer {
     constexpr explicit ConstBuffer(u32 max_offset_, bool is_indirect_)
@@ -64,16 +64,68 @@ struct GlobalMemoryUsage {
     bool is_written{};
 };
 
+class ShaderFunctionIR final {
+public:
+    explicit ShaderFunctionIR(std::map<u32, NodeBlock>&& basic_blocks_, bool disable_flow_stack_,
+                              u32 id_, u32 coverage_begin_, u32 coverage_end_)
+        : basic_blocks{std::move(basic_blocks_)}, decompiled{false},
+          disable_flow_stack{disable_flow_stack_}, id{id_}, coverage_begin{coverage_begin_},
+          coverage_end{coverage_end_} {}
+    explicit ShaderFunctionIR(ASTManager&& program_manager_, u32 id_, u32 coverage_begin_,
+                              u32 coverage_end_)
+        : program_manager{std::move(program_manager_)}, decompiled{true}, disable_flow_stack{true},
+          id{id_}, coverage_begin{coverage_begin_}, coverage_end{coverage_end_} {}
+
+    const std::map<u32, NodeBlock>& GetBasicBlocks() const {
+        return basic_blocks;
+    }
+
+    [[nodiscard]] bool IsFlowStackDisabled() const {
+        return disable_flow_stack;
+    }
+
+    [[nodiscard]] bool IsDecompiled() const {
+        return decompiled;
+    }
+
+    const ASTManager& GetASTManager() const {
+        return program_manager;
+    }
+
+    [[nodiscard]] ASTNode GetASTProgram() const {
+        return program_manager.GetProgram();
+    }
+
+    [[nodiscard]] u32 GetASTNumVariables() const {
+        return program_manager.GetVariables();
+    }
+
+    [[nodiscard]] bool IsMain() const {
+        return id == 0;
+    }
+
+    [[nodiscard]] u32 GetId() const {
+        return id;
+    }
+
+private:
+    std::map<u32, NodeBlock> basic_blocks;
+    ASTManager program_manager{true, true};
+
+    bool decompiled{};
+    bool disable_flow_stack{};
+    u32 id{};
+
+    u32 coverage_begin{};
+    u32 coverage_end{};
+};
+
 class ShaderIR final {
 public:
     explicit ShaderIR(const ProgramCode& program_code_, u32 main_offset_,
                       CompilerSettings settings_, Registry& registry_);
     ~ShaderIR();
 
-    const std::map<u32, NodeBlock>& GetBasicBlocks() const {
-        return basic_blocks;
-    }
-
     const std::set<u32>& GetRegisters() const {
         return used_registers;
     }
@@ -155,26 +207,6 @@ public:
         return header;
     }
 
-    bool IsFlowStackDisabled() const {
-        return disable_flow_stack;
-    }
-
-    bool IsDecompiled() const {
-        return decompiled;
-    }
-
-    const ASTManager& GetASTManager() const {
-        return program_manager;
-    }
-
-    ASTNode GetASTProgram() const {
-        return program_manager.GetProgram();
-    }
-
-    u32 GetASTNumVariables() const {
-        return program_manager.GetVariables();
-    }
-
     u32 ConvertAddressToNvidiaSpace(u32 address) const {
         return (address - main_offset) * static_cast<u32>(sizeof(Tegra::Shader::Instruction));
     }
@@ -190,7 +222,16 @@ public:
         return num_custom_variables;
     }
 
+    std::shared_ptr<ShaderFunctionIR> GetMainFunction() const {
+        return main_function;
+    }
+
+    const std::vector<std::shared_ptr<ShaderFunctionIR>>& GetSubFunctions() const {
+        return subfunctions;
+    }
+
 private:
+    friend class ExprDecoder;
     friend class ASTDecoder;
 
     struct SamplerInfo {
@@ -453,6 +494,10 @@ private:
     std::vector<Node> amend_code;
     u32 num_custom_variables{};
 
+    std::shared_ptr<ShaderFunctionIR> main_function;
+    std::vector<std::shared_ptr<ShaderFunctionIR>> subfunctions;
+    std::unordered_map<u32, u32> func_map;
+
     std::set<u32> used_registers;
     std::set<Tegra::Shader::Pred> used_predicates;
     std::set<Tegra::Shader::Attribute::Index> used_input_attributes;

src/video_core/texture_cache/image_base.cpp

@@ -69,6 +69,9 @@ ImageBase::ImageBase(const ImageInfo& info_, GPUVAddr gpu_addr_, VAddr cpu_addr_
     }
 }
 
+ImageMapView::ImageMapView(GPUVAddr gpu_addr_, VAddr cpu_addr_, size_t size_, ImageId image_id_)
+    : gpu_addr{gpu_addr_}, cpu_addr{cpu_addr_}, size{size_}, image_id{image_id_} {}
+
 std::optional<SubresourceBase> ImageBase::TryFindBase(GPUVAddr other_addr) const noexcept {
     if (other_addr < gpu_addr) {
         // Subresource address can't be lower than the base
@@ -82,7 +85,7 @@ std::optional<SubresourceBase> ImageBase::TryFindBase(GPUVAddr other_addr) const
     if (info.type != ImageType::e3D) {
         const auto [layer, mip_offset] = LayerMipOffset(diff, info.layer_stride);
         const auto end = mip_level_offsets.begin() + info.resources.levels;
-        const auto it = std::find(mip_level_offsets.begin(), end, mip_offset);
+        const auto it = std::find(mip_level_offsets.begin(), end, static_cast<u32>(mip_offset));
         if (layer > info.resources.layers || it == end) {
             return std::nullopt;
         }

src/video_core/texture_cache/image_base.h

@@ -25,12 +25,14 @@ enum class ImageFlagBits : u32 {
     Strong = 1 << 5,      ///< Exists in the image table, the dimensions are can be trusted
     Registered = 1 << 6,  ///< True when the image is registered
     Picked = 1 << 7,      ///< Temporary flag to mark the image as picked
+    Remapped = 1 << 8,    ///< Image has been remapped.
+    Sparse = 1 << 9,      ///< Image has non continous submemory.
 
     // Garbage Collection Flags
-    BadOverlap = 1 << 8, ///< This image overlaps other but doesn't fit, has higher
-                         ///< garbage collection priority
-    Alias = 1 << 9,      ///< This image has aliases and has priority on garbage
-                         ///< collection
+    BadOverlap = 1 << 10, ///< This image overlaps other but doesn't fit, has higher
+                          ///< garbage collection priority
+    Alias = 1 << 11,      ///< This image has aliases and has priority on garbage
+                          ///< collection
 };
 DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits)
 
@@ -57,6 +59,12 @@ struct ImageBase {
         return cpu_addr < overlap_end && overlap_cpu_addr < cpu_addr_end;
     }
 
+    [[nodiscard]] bool OverlapsGPU(GPUVAddr overlap_gpu_addr, size_t overlap_size) const noexcept {
+        const VAddr overlap_end = overlap_gpu_addr + overlap_size;
+        const GPUVAddr gpu_addr_end = gpu_addr + guest_size_bytes;
+        return gpu_addr < overlap_end && overlap_gpu_addr < gpu_addr_end;
+    }
+
     void CheckBadOverlapState();
     void CheckAliasState();
 
@@ -84,6 +92,29 @@ struct ImageBase {
 
     std::vector<AliasedImage> aliased_images;
     std::vector<ImageId> overlapping_images;
+    ImageMapId map_view_id{};
+};
+
+struct ImageMapView {
+    explicit ImageMapView(GPUVAddr gpu_addr, VAddr cpu_addr, size_t size, ImageId image_id);
+
+    [[nodiscard]] bool Overlaps(VAddr overlap_cpu_addr, size_t overlap_size) const noexcept {
+        const VAddr overlap_end = overlap_cpu_addr + overlap_size;
+        const VAddr cpu_addr_end = cpu_addr + size;
+        return cpu_addr < overlap_end && overlap_cpu_addr < cpu_addr_end;
+    }
+
+    [[nodiscard]] bool OverlapsGPU(GPUVAddr overlap_gpu_addr, size_t overlap_size) const noexcept {
+        const GPUVAddr overlap_end = overlap_gpu_addr + overlap_size;
+        const GPUVAddr gpu_addr_end = gpu_addr + size;
+        return gpu_addr < overlap_end && overlap_gpu_addr < gpu_addr_end;
+    }
+
+    GPUVAddr gpu_addr;
+    VAddr cpu_addr;
+    size_t size;
+    ImageId image_id;
+    bool picked{};
 };
 
 struct ImageAllocBase {

src/video_core/texture_cache/texture_cache.h

@@ -13,6 +13,7 @@
 #include <span>
 #include <type_traits>
 #include <unordered_map>
+#include <unordered_set>
 #include <utility>
 #include <vector>
 
@@ -152,6 +153,9 @@ public:
     /// Remove images in a region
     void UnmapMemory(VAddr cpu_addr, size_t size);
 
+    /// Remove images in a region
+    void UnmapGPUMemory(GPUVAddr gpu_addr, size_t size);
+
     /// Blit an image with the given parameters
     void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
                    const Tegra::Engines::Fermi2D::Surface& src,
@@ -190,7 +194,22 @@ public:
 private:
     /// Iterate over all page indices in a range
     template <typename Func>
-    static void ForEachPage(VAddr addr, size_t size, Func&& func) {
+    static void ForEachCPUPage(VAddr addr, size_t size, Func&& func) {
+        static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
+        const u64 page_end = (addr + size - 1) >> PAGE_BITS;
+        for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) {
+            if constexpr (RETURNS_BOOL) {
+                if (func(page)) {
+                    break;
+                }
+            } else {
+                func(page);
+            }
+        }
+    }
+
+    template <typename Func>
+    static void ForEachGPUPage(GPUVAddr addr, size_t size, Func&& func) {
         static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
         const u64 page_end = (addr + size - 1) >> PAGE_BITS;
         for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) {
@@ -220,7 +239,7 @@ private:
     FramebufferId GetFramebufferId(const RenderTargets& key);
 
     /// Refresh the contents (pixel data) of an image
-    void RefreshContents(Image& image);
+    void RefreshContents(Image& image, ImageId image_id);
 
     /// Upload data from guest to an image
     template <typename StagingBuffer>
@@ -269,6 +288,16 @@ private:
     template <typename Func>
     void ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& func);
 
+    template <typename Func>
+    void ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Func&& func);
+
+    template <typename Func>
+    void ForEachSparseImageInRegion(GPUVAddr gpu_addr, size_t size, Func&& func);
+
+    /// Iterates over all the images in a region calling func
+    template <typename Func>
+    void ForEachSparseSegment(ImageBase& image, Func&& func);
+
     /// Find or create an image view in the given image with the passed parameters
     [[nodiscard]] ImageViewId FindOrEmplaceImageView(ImageId image_id, const ImageViewInfo& info);
 
@@ -279,10 +308,10 @@ private:
     void UnregisterImage(ImageId image);
 
     /// Track CPU reads and writes for image
-    void TrackImage(ImageBase& image);
+    void TrackImage(ImageBase& image, ImageId image_id);
 
     /// Stop tracking CPU reads and writes for image
-    void UntrackImage(ImageBase& image);
+    void UntrackImage(ImageBase& image, ImageId image_id);
 
     /// Delete image from the cache
     void DeleteImage(ImageId image);
@@ -340,7 +369,13 @@ private:
     std::unordered_map<TSCEntry, SamplerId> samplers;
     std::unordered_map<RenderTargets, FramebufferId> framebuffers;
 
-    std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> page_table;
+    std::unordered_map<u64, std::vector<ImageMapId>, IdentityHash<u64>> page_table;
+    std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> gpu_page_table;
+    std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> sparse_page_table;
+
+    std::unordered_map<ImageId, std::vector<ImageViewId>> sparse_views;
+
+    VAddr virtual_invalid_space{};
 
     bool has_deleted_images = false;
     u64 total_used_memory = 0;
@@ -349,6 +384,7 @@ private:
     u64 critical_memory;
 
     SlotVector<Image> slot_images;
+    SlotVector<ImageMapView> slot_map_views;
     SlotVector<ImageView> slot_image_views;
     SlotVector<ImageAlloc> slot_image_allocs;
     SlotVector<Sampler> slot_samplers;
@@ -459,7 +495,7 @@ void TextureCache<P>::RunGarbageCollector() {
                 }
             }
             if (True(image->flags & ImageFlagBits::Tracked)) {
-                UntrackImage(*image);
+                UntrackImage(*image, image_id);
             }
             UnregisterImage(image_id);
             DeleteImage(image_id);
@@ -658,7 +694,9 @@ void TextureCache<P>::WriteMemory(VAddr cpu_addr, size_t size) {
             return;
         }
         image.flags |= ImageFlagBits::CpuModified;
-        UntrackImage(image);
+        if (True(image.flags & ImageFlagBits::Tracked)) {
+            UntrackImage(image, image_id);
+        }
     });
 }
 
@@ -695,13 +733,30 @@ void TextureCache<P>::UnmapMemory(VAddr cpu_addr, size_t size) {
     for (const ImageId id : deleted_images) {
         Image& image = slot_images[id];
         if (True(image.flags & ImageFlagBits::Tracked)) {
-            UntrackImage(image);
+            UntrackImage(image, id);
         }
         UnregisterImage(id);
         DeleteImage(id);
     }
 }
 
+template <class P>
+void TextureCache<P>::UnmapGPUMemory(GPUVAddr gpu_addr, size_t size) {
+    std::vector<ImageId> deleted_images;
+    ForEachImageInRegionGPU(gpu_addr, size,
+                            [&](ImageId id, Image&) { deleted_images.push_back(id); });
+    for (const ImageId id : deleted_images) {
+        Image& image = slot_images[id];
+        if (True(image.flags & ImageFlagBits::Remapped)) {
+            continue;
+        }
+        image.flags |= ImageFlagBits::Remapped;
+        if (True(image.flags & ImageFlagBits::Tracked)) {
+            UntrackImage(image, id);
+        }
+    }
+}
+
 template <class P>
 void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
                                 const Tegra::Engines::Fermi2D::Surface& src,
@@ -833,9 +888,10 @@ typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(VAddr cpu_ad
     if (it == page_table.end()) {
         return nullptr;
     }
-    const auto& image_ids = it->second;
-    for (const ImageId image_id : image_ids) {
-        const ImageBase& image = slot_images[image_id];
+    const auto& image_map_ids = it->second;
+    for (const ImageMapId map_id : image_map_ids) {
+        const ImageMapView& map = slot_map_views[map_id];
+        const ImageBase& image = slot_images[map.image_id];
         if (image.cpu_addr != cpu_addr) {
             continue;
         }
@@ -915,13 +971,13 @@ bool TextureCache<P>::IsRegionGpuModified(VAddr addr, size_t size) {
 }
 
 template <class P>
-void TextureCache<P>::RefreshContents(Image& image) {
+void TextureCache<P>::RefreshContents(Image& image, ImageId image_id) {
     if (False(image.flags & ImageFlagBits::CpuModified)) {
         // Only upload modified images
         return;
     }
     image.flags &= ~ImageFlagBits::CpuModified;
-    TrackImage(image);
+    TrackImage(image, image_id);
 
     if (image.info.num_samples > 1) {
         LOG_WARNING(HW_GPU, "MSAA image uploads are not implemented");
@@ -958,7 +1014,7 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging)
 
 template <class P>
 ImageViewId TextureCache<P>::FindImageView(const TICEntry& config) {
-    if (!IsValidAddress(gpu_memory, config)) {
+    if (!IsValidEntry(gpu_memory, config)) {
         return NULL_IMAGE_VIEW_ID;
     }
     const auto [pair, is_new] = image_views.try_emplace(config);
@@ -1000,14 +1056,20 @@ ImageId TextureCache<P>::FindOrInsertImage(const ImageInfo& info, GPUVAddr gpu_a
 template <class P>
 ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
                                    RelaxedOptions options) {
-    const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
+    std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
     if (!cpu_addr) {
-        return ImageId{};
+        cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr, CalculateGuestSizeInBytes(info));
+        if (!cpu_addr) {
+            return ImageId{};
+        }
     }
     const bool broken_views = runtime.HasBrokenTextureViewFormats();
     const bool native_bgr = runtime.HasNativeBgr();
     ImageId image_id;
     const auto lambda = [&](ImageId existing_image_id, ImageBase& existing_image) {
+        if (True(existing_image.flags & ImageFlagBits::Remapped)) {
+            return false;
+        }
         if (info.type == ImageType::Linear || existing_image.info.type == ImageType::Linear) {
             const bool strict_size = False(options & RelaxedOptions::Size) &&
                                      True(existing_image.flags & ImageFlagBits::Strong);
@@ -1033,7 +1095,16 @@ ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
 template <class P>
 ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
                                      RelaxedOptions options) {
-    const std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
+    std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
+    if (!cpu_addr) {
+        const auto size = CalculateGuestSizeInBytes(info);
+        cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr, size);
+        if (!cpu_addr) {
+            const VAddr fake_addr = ~(1ULL << 40ULL) + virtual_invalid_space;
+            virtual_invalid_space += Common::AlignUp(size, 32);
+            cpu_addr = std::optional<VAddr>(fake_addr);
+        }
+    }
     ASSERT_MSG(cpu_addr, "Tried to insert an image to an invalid gpu_addr=0x{:x}", gpu_addr);
     const ImageId image_id = JoinImages(info, gpu_addr, *cpu_addr);
     const Image& image = slot_images[image_id];
@@ -1053,10 +1124,16 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
     const bool broken_views = runtime.HasBrokenTextureViewFormats();
     const bool native_bgr = runtime.HasNativeBgr();
     std::vector<ImageId> overlap_ids;
+    std::unordered_set<ImageId> overlaps_found;
     std::vector<ImageId> left_aliased_ids;
     std::vector<ImageId> right_aliased_ids;
+    std::unordered_set<ImageId> ignore_textures;
     std::vector<ImageId> bad_overlap_ids;
-    ForEachImageInRegion(cpu_addr, size_bytes, [&](ImageId overlap_id, ImageBase& overlap) {
+    const auto region_check = [&](ImageId overlap_id, ImageBase& overlap) {
+        if (True(overlap.flags & ImageFlagBits::Remapped)) {
+            ignore_textures.insert(overlap_id);
+            return;
+        }
         if (info.type == ImageType::Linear) {
             if (info.pitch == overlap.info.pitch && gpu_addr == overlap.gpu_addr) {
                 // Alias linear images with the same pitch
@@ -1064,6 +1141,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
             }
             return;
         }
+        overlaps_found.insert(overlap_id);
         static constexpr bool strict_size = true;
         const std::optional<OverlapResult> solution = ResolveOverlap(
             new_info, gpu_addr, cpu_addr, overlap, strict_size, broken_views, native_bgr);
@@ -1087,12 +1165,40 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
             bad_overlap_ids.push_back(overlap_id);
             overlap.flags |= ImageFlagBits::BadOverlap;
         }
-    });
+    };
+    ForEachImageInRegion(cpu_addr, size_bytes, region_check);
+    const auto region_check_gpu = [&](ImageId overlap_id, ImageBase& overlap) {
+        if (!overlaps_found.contains(overlap_id)) {
+            if (True(overlap.flags & ImageFlagBits::Remapped)) {
+                ignore_textures.insert(overlap_id);
+            }
+            if (overlap.gpu_addr == gpu_addr && overlap.guest_size_bytes == size_bytes) {
+                ignore_textures.insert(overlap_id);
+            }
+        }
+    };
+    ForEachSparseImageInRegion(gpu_addr, size_bytes, region_check_gpu);
     const ImageId new_image_id = slot_images.insert(runtime, new_info, gpu_addr, cpu_addr);
     Image& new_image = slot_images[new_image_id];
 
+    if (!gpu_memory.IsContinousRange(new_image.gpu_addr, new_image.guest_size_bytes)) {
+        new_image.flags |= ImageFlagBits::Sparse;
+    }
+
+    for (const ImageId overlap_id : ignore_textures) {
+        Image& overlap = slot_images[overlap_id];
+        if (True(overlap.flags & ImageFlagBits::GpuModified)) {
+            UNIMPLEMENTED();
+        }
+        if (True(overlap.flags & ImageFlagBits::Tracked)) {
+            UntrackImage(overlap, overlap_id);
+        }
+        UnregisterImage(overlap_id);
+        DeleteImage(overlap_id);
+    }
+
     // TODO: Only upload what we need
-    RefreshContents(new_image);
+    RefreshContents(new_image, new_image_id);
 
     for (const ImageId overlap_id : overlap_ids) {
         Image& overlap = slot_images[overlap_id];
@@ -1104,7 +1210,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
             runtime.CopyImage(new_image, overlap, copies);
         }
         if (True(overlap.flags & ImageFlagBits::Tracked)) {
-            UntrackImage(overlap);
+            UntrackImage(overlap, overlap_id);
         }
         UnregisterImage(overlap_id);
         DeleteImage(overlap_id);
@@ -1239,7 +1345,8 @@ void TextureCache<P>::ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& f
     using FuncReturn = typename std::invoke_result<Func, ImageId, Image&>::type;
     static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>;
     boost::container::small_vector<ImageId, 32> images;
-    ForEachPage(cpu_addr, size, [this, &images, cpu_addr, size, func](u64 page) {
+    boost::container::small_vector<ImageMapId, 32> maps;
+    ForEachCPUPage(cpu_addr, size, [this, &images, &maps, cpu_addr, size, func](u64 page) {
         const auto it = page_table.find(page);
         if (it == page_table.end()) {
             if constexpr (BOOL_BREAK) {
@@ -1248,12 +1355,63 @@ void TextureCache<P>::ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& f
                 return;
             }
         }
+        for (const ImageMapId map_id : it->second) {
+            ImageMapView& map = slot_map_views[map_id];
+            if (map.picked) {
+                continue;
+            }
+            if (!map.Overlaps(cpu_addr, size)) {
+                continue;
+            }
+            map.picked = true;
+            maps.push_back(map_id);
+            Image& image = slot_images[map.image_id];
+            if (True(image.flags & ImageFlagBits::Picked)) {
+                continue;
+            }
+            image.flags |= ImageFlagBits::Picked;
+            images.push_back(map.image_id);
+            if constexpr (BOOL_BREAK) {
+                if (func(map.image_id, image)) {
+                    return true;
+                }
+            } else {
+                func(map.image_id, image);
+            }
+        }
+        if constexpr (BOOL_BREAK) {
+            return false;
+        }
+    });
+    for (const ImageId image_id : images) {
+        slot_images[image_id].flags &= ~ImageFlagBits::Picked;
+    }
+    for (const ImageMapId map_id : maps) {
+        slot_map_views[map_id].picked = false;
+    }
+}
+
+template <class P>
+template <typename Func>
+void TextureCache<P>::ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Func&& func) {
+    using FuncReturn = typename std::invoke_result<Func, ImageId, Image&>::type;
+    static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>;
+    boost::container::small_vector<ImageId, 8> images;
+    ForEachGPUPage(gpu_addr, size, [this, &images, gpu_addr, size, func](u64 page) {
+        const auto it = gpu_page_table.find(page);
+        if (it == gpu_page_table.end()) {
+            if constexpr (BOOL_BREAK) {
+                return false;
+            } else {
+                return;
+            }
+        }
         for (const ImageId image_id : it->second) {
             Image& image = slot_images[image_id];
             if (True(image.flags & ImageFlagBits::Picked)) {
                 continue;
             }
-            if (!image.Overlaps(cpu_addr, size)) {
+            if (!image.OverlapsGPU(gpu_addr, size)) {
                 continue;
             }
             image.flags |= ImageFlagBits::Picked;
@@ -1275,6 +1433,69 @@ void TextureCache<P>::ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& f
     }
 }
 
+template <class P>
+template <typename Func>
+void TextureCache<P>::ForEachSparseImageInRegion(GPUVAddr gpu_addr, size_t size, Func&& func) {
+    using FuncReturn = typename std::invoke_result<Func, ImageId, Image&>::type;
+    static constexpr bool BOOL_BREAK = std::is_same_v<FuncReturn, bool>;
+    boost::container::small_vector<ImageId, 8> images;
+    ForEachGPUPage(gpu_addr, size, [this, &images, gpu_addr, size, func](u64 page) {
+        const auto it = sparse_page_table.find(page);
+        if (it == sparse_page_table.end()) {
+            if constexpr (BOOL_BREAK) {
+                return false;
+            } else {
+                return;
+            }
+        }
+        for (const ImageId image_id : it->second) {
+            Image& image = slot_images[image_id];
+            if (True(image.flags & ImageFlagBits::Picked)) {
+                continue;
+            }
+            if (!image.OverlapsGPU(gpu_addr, size)) {
+                continue;
+            }
+            image.flags |= ImageFlagBits::Picked;
+            images.push_back(image_id);
+            if constexpr (BOOL_BREAK) {
+                if (func(image_id, image)) {
+                    return true;
+                }
+            } else {
+                func(image_id, image);
+            }
+        }
+        if constexpr (BOOL_BREAK) {
+            return false;
+        }
+    });
+    for (const ImageId image_id : images) {
+        slot_images[image_id].flags &= ~ImageFlagBits::Picked;
+    }
+}
+
+template <class P>
+template <typename Func>
+void TextureCache<P>::ForEachSparseSegment(ImageBase& image, Func&& func) {
+    using FuncReturn = typename std::invoke_result<Func, GPUVAddr, VAddr, size_t>::type;
+    static constexpr bool RETURNS_BOOL = std::is_same_v<FuncReturn, bool>;
+    const auto segments = gpu_memory.GetSubmappedRange(image.gpu_addr, image.guest_size_bytes);
+    for (auto& segment : segments) {
+        const auto gpu_addr = segment.first;
+        const auto size = segment.second;
+        std::optional<VAddr> cpu_addr = gpu_memory.GpuToCpuAddress(gpu_addr);
+        ASSERT(cpu_addr);
+        if constexpr (RETURNS_BOOL) {
+            if (func(gpu_addr, *cpu_addr, size)) {
+                break;
+            }
+        } else {
+            func(gpu_addr, *cpu_addr, size);
+        }
+    }
+}
+
 template <class P>
 ImageViewId TextureCache<P>::FindOrEmplaceImageView(ImageId image_id, const ImageViewInfo& info) {
     Image& image = slot_images[image_id];
@@ -1292,8 +1513,6 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
     ASSERT_MSG(False(image.flags & ImageFlagBits::Registered),
                "Trying to register an already registered image");
     image.flags |= ImageFlagBits::Registered;
-    ForEachPage(image.cpu_addr, image.guest_size_bytes,
-                [this, image_id](u64 page) { page_table[page].push_back(image_id); });
     u64 tentative_size = std::max(image.guest_size_bytes, image.unswizzled_size_bytes);
     if ((IsPixelFormatASTC(image.info.format) &&
          True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
@@ -1301,6 +1520,27 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
         tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
     }
     total_used_memory += Common::AlignUp(tentative_size, 1024);
+    ForEachGPUPage(image.gpu_addr, image.guest_size_bytes,
+                   [this, image_id](u64 page) { gpu_page_table[page].push_back(image_id); });
+    if (False(image.flags & ImageFlagBits::Sparse)) {
+        auto map_id =
+            slot_map_views.insert(image.gpu_addr, image.cpu_addr, image.guest_size_bytes, image_id);
+        ForEachCPUPage(image.cpu_addr, image.guest_size_bytes,
+                       [this, map_id](u64 page) { page_table[page].push_back(map_id); });
+        image.map_view_id = map_id;
+        return;
+    }
+    std::vector<ImageViewId> sparse_maps{};
+    ForEachSparseSegment(
+        image, [this, image_id, &sparse_maps](GPUVAddr gpu_addr, VAddr cpu_addr, size_t size) {
+            auto map_id = slot_map_views.insert(gpu_addr, cpu_addr, size, image_id);
+            ForEachCPUPage(cpu_addr, size,
+                           [this, map_id](u64 page) { page_table[page].push_back(map_id); });
+            sparse_maps.push_back(map_id);
+        });
+    sparse_views.emplace(image_id, std::move(sparse_maps));
+    ForEachGPUPage(image.gpu_addr, image.guest_size_bytes,
+                   [this, image_id](u64 page) { sparse_page_table[page].push_back(image_id); });
 }
 
 template <class P>
@@ -1317,34 +1557,125 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
         tentative_size = EstimatedDecompressedSize(tentative_size, image.info.format);
     }
     total_used_memory -= Common::AlignUp(tentative_size, 1024);
-    ForEachPage(image.cpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {
-        const auto page_it = page_table.find(page);
-        if (page_it == page_table.end()) {
-            UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS);
-            return;
-        }
-        std::vector<ImageId>& image_ids = page_it->second;
-        const auto vector_it = std::ranges::find(image_ids, image_id);
-        if (vector_it == image_ids.end()) {
-            UNREACHABLE_MSG("Unregistering unregistered image in page=0x{:x}", page << PAGE_BITS);
-            return;
-        }
-        image_ids.erase(vector_it);
+    const auto& clear_page_table =
+        [this, image_id](
+            u64 page,
+            std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>>& selected_page_table) {
+            const auto page_it = selected_page_table.find(page);
+            if (page_it == selected_page_table.end()) {
+                UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS);
+                return;
+            }
+            std::vector<ImageId>& image_ids = page_it->second;
+            const auto vector_it = std::ranges::find(image_ids, image_id);
+            if (vector_it == image_ids.end()) {
+                UNREACHABLE_MSG("Unregistering unregistered image in page=0x{:x}",
+                                page << PAGE_BITS);
+                return;
+            }
+            image_ids.erase(vector_it);
+        };
+    ForEachGPUPage(image.gpu_addr, image.guest_size_bytes,
+                   [this, &clear_page_table](u64 page) { clear_page_table(page, gpu_page_table); });
+    if (False(image.flags & ImageFlagBits::Sparse)) {
+        const auto map_id = image.map_view_id;
+        ForEachCPUPage(image.cpu_addr, image.guest_size_bytes, [this, map_id](u64 page) {
+            const auto page_it = page_table.find(page);
+            if (page_it == page_table.end()) {
+                UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS);
+                return;
+            }
+            std::vector<ImageMapId>& image_map_ids = page_it->second;
+            const auto vector_it = std::ranges::find(image_map_ids, map_id);
+            if (vector_it == image_map_ids.end()) {
+                UNREACHABLE_MSG("Unregistering unregistered image in page=0x{:x}",
+                                page << PAGE_BITS);
+                return;
+            }
+            image_map_ids.erase(vector_it);
+        });
+        slot_map_views.erase(map_id);
+        return;
+    }
+    ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, &clear_page_table](u64 page) {
+        clear_page_table(page, sparse_page_table);
     });
+    auto it = sparse_views.find(image_id);
+    ASSERT(it != sparse_views.end());
+    auto& sparse_maps = it->second;
+    for (auto& map_view_id : sparse_maps) {
+        const auto& map_range = slot_map_views[map_view_id];
+        const VAddr cpu_addr = map_range.cpu_addr;
+        const std::size_t size = map_range.size;
+        ForEachCPUPage(cpu_addr, size, [this, image_id](u64 page) {
+            const auto page_it = page_table.find(page);
+            if (page_it == page_table.end()) {
+                UNREACHABLE_MSG("Unregistering unregistered page=0x{:x}", page << PAGE_BITS);
+                return;
+            }
+            std::vector<ImageMapId>& image_map_ids = page_it->second;
+            auto vector_it = image_map_ids.begin();
+            while (vector_it != image_map_ids.end()) {
+                ImageMapView& map = slot_map_views[*vector_it];
+                if (map.image_id != image_id) {
+                    vector_it++;
+                    continue;
+                }
+                if (!map.picked) {
+                    map.picked = true;
+                }
+                vector_it = image_map_ids.erase(vector_it);
+            }
+        });
+        slot_map_views.erase(map_view_id);
+    }
+    sparse_views.erase(it);
 }
 
 template <class P>
-void TextureCache<P>::TrackImage(ImageBase& image) {
+void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) {
     ASSERT(False(image.flags & ImageFlagBits::Tracked));
     image.flags |= ImageFlagBits::Tracked;
-    rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1);
+    if (False(image.flags & ImageFlagBits::Sparse)) {
+        rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1);
+        return;
+    }
+    if (True(image.flags & ImageFlagBits::Registered)) {
+        auto it = sparse_views.find(image_id);
+        ASSERT(it != sparse_views.end());
+        auto& sparse_maps = it->second;
+        for (auto& map_view_id : sparse_maps) {
+            const auto& map = slot_map_views[map_view_id];
+            const VAddr cpu_addr = map.cpu_addr;
+            const std::size_t size = map.size;
+            rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1);
+        }
+        return;
+    }
+    ForEachSparseSegment(image,
+                         [this]([[maybe_unused]] GPUVAddr gpu_addr, VAddr cpu_addr, size_t size) {
+                             rasterizer.UpdatePagesCachedCount(cpu_addr, size, 1);
+                         });
 }
 
 template <class P>
-void TextureCache<P>::UntrackImage(ImageBase& image) {
+void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) {
     ASSERT(True(image.flags & ImageFlagBits::Tracked));
     image.flags &= ~ImageFlagBits::Tracked;
-    rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, -1);
+    if (False(image.flags & ImageFlagBits::Sparse)) {
+        rasterizer.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, -1);
+        return;
+    }
+    ASSERT(True(image.flags & ImageFlagBits::Registered));
+    auto it = sparse_views.find(image_id);
+    ASSERT(it != sparse_views.end());
+    auto& sparse_maps = it->second;
+    for (auto& map_view_id : sparse_maps) {
+        const auto& map = slot_map_views[map_view_id];
+        const VAddr cpu_addr = map.cpu_addr;
+        const std::size_t size = map.size;
+        rasterizer.UpdatePagesCachedCount(cpu_addr, size, -1);
+    }
 }
 
 template <class P>
@@ -1486,10 +1817,10 @@ void TextureCache<P>::PrepareImage(ImageId image_id, bool is_modification, bool
     if (invalidate) {
         image.flags &= ~(ImageFlagBits::CpuModified | ImageFlagBits::GpuModified);
         if (False(image.flags & ImageFlagBits::Tracked)) {
-            TrackImage(image);
+            TrackImage(image, image_id);
         }
     } else {
-        RefreshContents(image);
+        RefreshContents(image, image_id);
         SynchronizeAliases(image_id);
     }
     if (is_modification) {

src/video_core/texture_cache/types.h

@@ -16,6 +16,7 @@ constexpr size_t MAX_MIP_LEVELS = 14;
 constexpr SlotId CORRUPT_ID{0xfffffffe};
 
 using ImageId = SlotId;
+using ImageMapId = SlotId;
 using ImageViewId = SlotId;
 using ImageAllocId = SlotId;
 using SamplerId = SlotId;

src/video_core/texture_cache/util.cpp

@@ -394,7 +394,7 @@ template <u32 GOB_EXTENT>
     const s32 mip_offset = diff % layer_stride;
     const std::array offsets = CalculateMipLevelOffsets(new_info);
     const auto end = offsets.begin() + new_info.resources.levels;
-    const auto it = std::find(offsets.begin(), end, mip_offset);
+    const auto it = std::find(offsets.begin(), end, static_cast<u32>(mip_offset));
     if (it == end) {
         // Mipmap is not aligned to any valid size
         return std::nullopt;
@@ -664,6 +664,16 @@ LevelArray CalculateMipLevelOffsets(const ImageInfo& info) noexcept {
     return offsets;
 }
 
+LevelArray CalculateMipLevelSizes(const ImageInfo& info) noexcept {
+    const u32 num_levels = info.resources.levels;
+    const LevelInfo level_info = MakeLevelInfo(info);
+    LevelArray sizes{};
+    for (u32 level = 0; level < num_levels; ++level) {
+        sizes[level] = CalculateLevelSize(level_info, level);
+    }
+    return sizes;
+}
+
 std::vector<u32> CalculateSliceOffsets(const ImageInfo& info) {
     ASSERT(info.type == ImageType::e3D);
     std::vector<u32> offsets;
@@ -776,14 +786,20 @@ std::vector<ImageCopy> MakeShrinkImageCopies(const ImageInfo& dst, const ImageIn
     return copies;
 }
 
-bool IsValidAddress(const Tegra::MemoryManager& gpu_memory, const TICEntry& config) {
-    if (config.Address() == 0) {
+bool IsValidEntry(const Tegra::MemoryManager& gpu_memory, const TICEntry& config) {
+    const GPUVAddr address = config.Address();
+    if (address == 0) {
         return false;
     }
-    if (config.Address() > (u64(1) << 48)) {
+    if (address > (1ULL << 48)) {
         return false;
     }
-    return gpu_memory.GpuToCpuAddress(config.Address()).has_value();
+    if (gpu_memory.GpuToCpuAddress(address).has_value()) {
+        return true;
+    }
+    const ImageInfo info{config};
+    const size_t guest_size_bytes = CalculateGuestSizeInBytes(info);
+    return gpu_memory.GpuToCpuAddress(address, guest_size_bytes).has_value();
 }
 
 std::vector<BufferImageCopy> UnswizzleImage(Tegra::MemoryManager& gpu_memory, GPUVAddr gpu_addr,

src/video_core/texture_cache/util.h

@@ -40,6 +40,8 @@ struct OverlapResult {
 
 [[nodiscard]] LevelArray CalculateMipLevelOffsets(const ImageInfo& info) noexcept;
 
+[[nodiscard]] LevelArray CalculateMipLevelSizes(const ImageInfo& info) noexcept;
+
 [[nodiscard]] std::vector<u32> CalculateSliceOffsets(const ImageInfo& info);
 
 [[nodiscard]] std::vector<SubresourceBase> CalculateSliceSubresources(const ImageInfo& info);
@@ -55,7 +57,7 @@ struct OverlapResult {
                                                            const ImageInfo& src,
                                                            SubresourceBase base);
 
-[[nodiscard]] bool IsValidAddress(const Tegra::MemoryManager& gpu_memory, const TICEntry& config);
+[[nodiscard]] bool IsValidEntry(const Tegra::MemoryManager& gpu_memory, const TICEntry& config);
 
 [[nodiscard]] std::vector<BufferImageCopy> UnswizzleImage(Tegra::MemoryManager& gpu_memory,
                                                           GPUVAddr gpu_addr, const ImageInfo& info,

src/video_core/textures/astc.cpp

@@ -1365,8 +1365,8 @@ static void DecompressBlock(std::span<const u8, 16> inBuf, const u32 blockWidth,
     // each partition.
 
     // Determine partitions, partition index, and color endpoint modes
-    s32 planeIdx = -1;
-    u32 partitionIndex;
+    u32 planeIdx{UINT32_MAX};
+    u32 partitionIndex{};
     u32 colorEndpointMode[4] = {0, 0, 0, 0};
 
     // Define color data.

src/yuzu/debugger/profiler.cpp

@@ -143,24 +143,25 @@ void MicroProfileWidget::hideEvent(QHideEvent* ev) {
 }
 
 void MicroProfileWidget::mouseMoveEvent(QMouseEvent* ev) {
-    MicroProfileMousePosition(ev->x() / x_scale, ev->y() / y_scale, 0);
+    MicroProfileMousePosition(ev->pos().x() / x_scale, ev->pos().y() / y_scale, 0);
     ev->accept();
 }
 
 void MicroProfileWidget::mousePressEvent(QMouseEvent* ev) {
-    MicroProfileMousePosition(ev->x() / x_scale, ev->y() / y_scale, 0);
+    MicroProfileMousePosition(ev->pos().x() / x_scale, ev->pos().y() / y_scale, 0);
     MicroProfileMouseButton(ev->buttons() & Qt::LeftButton, ev->buttons() & Qt::RightButton);
     ev->accept();
 }
 
 void MicroProfileWidget::mouseReleaseEvent(QMouseEvent* ev) {
-    MicroProfileMousePosition(ev->x() / x_scale, ev->y() / y_scale, 0);
+    MicroProfileMousePosition(ev->pos().x() / x_scale, ev->pos().y() / y_scale, 0);
     MicroProfileMouseButton(ev->buttons() & Qt::LeftButton, ev->buttons() & Qt::RightButton);
     ev->accept();
 }
 
 void MicroProfileWidget::wheelEvent(QWheelEvent* ev) {
-    MicroProfileMousePosition(ev->x() / x_scale, ev->y() / y_scale, ev->delta() / 120);
+    MicroProfileMousePosition(ev->pos().x() / x_scale, ev->pos().y() / y_scale,
+                              ev->angleDelta().y() / 120);
     ev->accept();
 }