Merge pull request #2866 from Morph1984/checkbox_fix

configure_input: Uncheck "Joycons Docked" when "Use Docked Mode" is checked

src/core/hle/service/acc/acc.cpp

@@ -31,6 +31,9 @@
 
 namespace Service::Account {
 
+constexpr ResultCode ERR_INVALID_BUFFER_SIZE{ErrorModule::Account, 30};
+constexpr ResultCode ERR_FAILED_SAVE_DATA{ErrorModule::Account, 100};
+
 static std::string GetImagePath(Common::UUID uuid) {
     return FileUtil::GetUserPath(FileUtil::UserPath::NANDDir) +
            "/system/save/8000000000000010/su/avators/" + uuid.FormatSwitch() + ".jpg";
@@ -41,20 +44,31 @@ static constexpr u32 SanitizeJPEGSize(std::size_t size) {
     return static_cast<u32>(std::min(size, max_jpeg_image_size));
 }
 
-class IProfile final : public ServiceFramework<IProfile> {
+class IProfileCommon : public ServiceFramework<IProfileCommon> {
 public:
-    explicit IProfile(Common::UUID user_id, ProfileManager& profile_manager)
-        : ServiceFramework("IProfile"), profile_manager(profile_manager), user_id(user_id) {
+    explicit IProfileCommon(const char* name, bool editor_commands, Common::UUID user_id,
+                            ProfileManager& profile_manager)
+        : ServiceFramework(name), profile_manager(profile_manager), user_id(user_id) {
         static const FunctionInfo functions[] = {
-            {0, &IProfile::Get, "Get"},
-            {1, &IProfile::GetBase, "GetBase"},
-            {10, &IProfile::GetImageSize, "GetImageSize"},
-            {11, &IProfile::LoadImage, "LoadImage"},
+            {0, &IProfileCommon::Get, "Get"},
+            {1, &IProfileCommon::GetBase, "GetBase"},
+            {10, &IProfileCommon::GetImageSize, "GetImageSize"},
+            {11, &IProfileCommon::LoadImage, "LoadImage"},
         };
+
         RegisterHandlers(functions);
+
+        if (editor_commands) {
+            static const FunctionInfo editor_functions[] = {
+                {100, &IProfileCommon::Store, "Store"},
+                {101, &IProfileCommon::StoreWithImage, "StoreWithImage"},
+            };
+
+            RegisterHandlers(editor_functions);
+        }
     }
 
-private:
+protected:
     void Get(Kernel::HLERequestContext& ctx) {
         LOG_INFO(Service_ACC, "called user_id={}", user_id.Format());
         ProfileBase profile_base{};
@@ -127,10 +141,91 @@ private:
         }
     }
 
-    const ProfileManager& profile_manager;
+    void Store(Kernel::HLERequestContext& ctx) {
+        IPC::RequestParser rp{ctx};
+        const auto base = rp.PopRaw<ProfileBase>();
+
+        const auto user_data = ctx.ReadBuffer();
+
+        LOG_DEBUG(Service_ACC, "called, username='{}', timestamp={:016X}, uuid={}",
+                  Common::StringFromFixedZeroTerminatedBuffer(
+                      reinterpret_cast<const char*>(base.username.data()), base.username.size()),
+                  base.timestamp, base.user_uuid.Format());
+
+        if (user_data.size() < sizeof(ProfileData)) {
+            LOG_ERROR(Service_ACC, "ProfileData buffer too small!");
+            IPC::ResponseBuilder rb{ctx, 2};
+            rb.Push(ERR_INVALID_BUFFER_SIZE);
+            return;
+        }
+
+        ProfileData data;
+        std::memcpy(&data, user_data.data(), sizeof(ProfileData));
+
+        if (!profile_manager.SetProfileBaseAndData(user_id, base, data)) {
+            LOG_ERROR(Service_ACC, "Failed to update profile data and base!");
+            IPC::ResponseBuilder rb{ctx, 2};
+            rb.Push(ERR_FAILED_SAVE_DATA);
+            return;
+        }
+
+        IPC::ResponseBuilder rb{ctx, 2};
+        rb.Push(RESULT_SUCCESS);
+    }
+
+    void StoreWithImage(Kernel::HLERequestContext& ctx) {
+        IPC::RequestParser rp{ctx};
+        const auto base = rp.PopRaw<ProfileBase>();
+
+        const auto user_data = ctx.ReadBuffer();
+        const auto image_data = ctx.ReadBuffer(1);
+
+        LOG_DEBUG(Service_ACC, "called, username='{}', timestamp={:016X}, uuid={}",
+                  Common::StringFromFixedZeroTerminatedBuffer(
+                      reinterpret_cast<const char*>(base.username.data()), base.username.size()),
+                  base.timestamp, base.user_uuid.Format());
+
+        if (user_data.size() < sizeof(ProfileData)) {
+            LOG_ERROR(Service_ACC, "ProfileData buffer too small!");
+            IPC::ResponseBuilder rb{ctx, 2};
+            rb.Push(ERR_INVALID_BUFFER_SIZE);
+            return;
+        }
+
+        ProfileData data;
+        std::memcpy(&data, user_data.data(), sizeof(ProfileData));
+
+        FileUtil::IOFile image(GetImagePath(user_id), "wb");
+
+        if (!image.IsOpen() || !image.Resize(image_data.size()) ||
+            image.WriteBytes(image_data.data(), image_data.size()) != image_data.size() ||
+            !profile_manager.SetProfileBaseAndData(user_id, base, data)) {
+            LOG_ERROR(Service_ACC, "Failed to update profile data, base, and image!");
+            IPC::ResponseBuilder rb{ctx, 2};
+            rb.Push(ERR_FAILED_SAVE_DATA);
+            return;
+        }
+
+        IPC::ResponseBuilder rb{ctx, 2};
+        rb.Push(RESULT_SUCCESS);
+    }
+
+    ProfileManager& profile_manager;
     Common::UUID user_id; ///< The user id this profile refers to.
 };
 
+class IProfile final : public IProfileCommon {
+public:
+    IProfile(Common::UUID user_id, ProfileManager& profile_manager)
+        : IProfileCommon("IProfile", false, user_id, profile_manager) {}
+};
+
+class IProfileEditor final : public IProfileCommon {
+public:
+    IProfileEditor(Common::UUID user_id, ProfileManager& profile_manager)
+        : IProfileCommon("IProfileEditor", true, user_id, profile_manager) {}
+};
+
 class IManagerForApplication final : public ServiceFramework<IManagerForApplication> {
 public:
     IManagerForApplication() : ServiceFramework("IManagerForApplication") {
@@ -322,6 +417,17 @@ void Module::Interface::IsUserAccountSwitchLocked(Kernel::HLERequestContext& ctx
     rb.Push(is_locked);
 }
 
+void Module::Interface::GetProfileEditor(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    Common::UUID user_id = rp.PopRaw<Common::UUID>();
+
+    LOG_DEBUG(Service_ACC, "called, user_id={}", user_id.Format());
+
+    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
+    rb.Push(RESULT_SUCCESS);
+    rb.PushIpcInterface<IProfileEditor>(user_id, *profile_manager);
+}
+
 void Module::Interface::TrySelectUserWithoutInteraction(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_ACC, "called");
     // A u8 is passed into this function which we can safely ignore. It's to determine if we have

src/core/hle/service/acc/acc.h

@@ -32,6 +32,7 @@ public:
         void IsUserRegistrationRequestPermitted(Kernel::HLERequestContext& ctx);
         void TrySelectUserWithoutInteraction(Kernel::HLERequestContext& ctx);
         void IsUserAccountSwitchLocked(Kernel::HLERequestContext& ctx);
+        void GetProfileEditor(Kernel::HLERequestContext& ctx);
 
     private:
         ResultCode InitializeApplicationInfoBase(u64 process_id);

src/core/hle/service/acc/acc_su.cpp

@@ -41,7 +41,7 @@ ACC_SU::ACC_SU(std::shared_ptr<Module> module, std::shared_ptr<ProfileManager> p
         {202, nullptr, "CancelUserRegistration"},
         {203, nullptr, "DeleteUser"},
         {204, nullptr, "SetUserPosition"},
-        {205, nullptr, "GetProfileEditor"},
+        {205, &ACC_SU::GetProfileEditor, "GetProfileEditor"},
         {206, nullptr, "CompleteUserRegistrationForcibly"},
         {210, nullptr, "CreateFloatingRegistrationRequest"},
         {230, nullptr, "AuthenticateServiceAsync"},

src/core/hle/service/acc/profile_manager.cpp

@@ -305,6 +305,17 @@ bool ProfileManager::SetProfileBase(UUID uuid, const ProfileBase& profile_new) {
     return true;
 }
 
+bool ProfileManager::SetProfileBaseAndData(Common::UUID uuid, const ProfileBase& profile_new,
+                                           const ProfileData& data_new) {
+    const auto index = GetUserIndex(uuid);
+    if (index.has_value() && SetProfileBase(uuid, profile_new)) {
+        profiles[*index].data = data_new;
+        return true;
+    }
+
+    return false;
+}
+
 void ProfileManager::ParseUserSaveFile() {
     FileUtil::IOFile save(FileUtil::GetUserPath(FileUtil::UserPath::NANDDir) +
                               ACC_SAVE_AVATORS_BASE_PATH + "profiles.dat",

src/core/hle/service/acc/profile_manager.h

@@ -91,6 +91,8 @@ public:
 
     bool RemoveUser(Common::UUID uuid);
     bool SetProfileBase(Common::UUID uuid, const ProfileBase& profile_new);
+    bool SetProfileBaseAndData(Common::UUID uuid, const ProfileBase& profile_new,
+                               const ProfileData& data_new);
 
 private:
     void ParseUserSaveFile();

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

@@ -146,8 +146,8 @@ u32 nvhost_gpu::SubmitGPFIFO(const std::vector<u8>& input, std::vector<u8>& outp
     }
     IoctlSubmitGpfifo params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmitGpfifo));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called, gpfifo={:X}, num_entries={:X}, flags={:X}",
-                params.address, params.num_entries, params.flags.raw);
+    LOG_TRACE(Service_NVDRV, "called, gpfifo={:X}, num_entries={:X}, flags={:X}", params.address,
+              params.num_entries, params.flags.raw);
 
     ASSERT_MSG(input.size() == sizeof(IoctlSubmitGpfifo) +
                                    params.num_entries * sizeof(Tegra::CommandListHeader),
@@ -179,8 +179,8 @@ u32 nvhost_gpu::KickoffPB(const std::vector<u8>& input, std::vector<u8>& output)
     }
     IoctlSubmitGpfifo params{};
     std::memcpy(&params, input.data(), sizeof(IoctlSubmitGpfifo));
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called, gpfifo={:X}, num_entries={:X}, flags={:X}",
-                params.address, params.num_entries, params.flags.raw);
+    LOG_TRACE(Service_NVDRV, "called, gpfifo={:X}, num_entries={:X}, flags={:X}", params.address,
+              params.num_entries, params.flags.raw);
 
     Tegra::CommandList entries(params.num_entries);
     Memory::ReadBlock(params.address, entries.data(),

src/video_core/engines/maxwell_3d.cpp

@@ -89,6 +89,9 @@ void Maxwell3D::InitializeRegisterDefaults() {
 
     // Commercial games seem to assume this value is enabled and nouveau sets this value manually.
     regs.rt_separate_frag_data = 1;
+
+    // Some games (like Super Mario Odyssey) assume that SRGB is enabled.
+    regs.framebuffer_srgb = 1;
 }
 
 #define DIRTY_REGS_POS(field_name) (offsetof(Maxwell3D::DirtyRegs, field_name))
@@ -329,6 +332,10 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         ProcessMacroBind(method_call.argument);
         break;
     }
+    case MAXWELL3D_REG_INDEX(firmware[4]): {
+        ProcessFirmwareCall4();
+        break;
+    }
     case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]):
     case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]):
     case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]):
@@ -419,6 +426,14 @@ void Maxwell3D::ProcessMacroBind(u32 data) {
     macro_positions[regs.macros.entry++] = data;
 }
 
+void Maxwell3D::ProcessFirmwareCall4() {
+    LOG_WARNING(HW_GPU, "(STUBBED) called");
+
+    // Firmware call 4 is a blob that changes some registers depending on its parameters.
+    // These registers don't affect emulation and so are stubbed by setting 0xd00 to 1.
+    regs.reg_array[0xd00] = 1;
+}
+
 void Maxwell3D::ProcessQueryGet() {
     const GPUVAddr sequence_address{regs.query.QueryAddress()};
     // Since the sequence address is given as a GPU VAddr, we have to convert it to an application
@@ -526,7 +541,7 @@ void Maxwell3D::ProcessSyncPoint() {
 }
 
 void Maxwell3D::DrawArrays() {
-    LOG_DEBUG(HW_GPU, "called, topology={}, count={}", static_cast<u32>(regs.draw.topology.Value()),
+    LOG_TRACE(HW_GPU, "called, topology={}, count={}", static_cast<u32>(regs.draw.topology.Value()),
               regs.vertex_buffer.count);
     ASSERT_MSG(!(regs.index_array.count && regs.vertex_buffer.count), "Both indexed and direct?");
 

src/video_core/engines/maxwell_3d.h

@@ -1089,7 +1089,9 @@ public:
                     INSERT_PADDING_WORDS(14);
                 } shader_config[MaxShaderProgram];
 
-                INSERT_PADDING_WORDS(0x80);
+                INSERT_PADDING_WORDS(0x60);
+
+                u32 firmware[0x20];
 
                 struct {
                     u32 cb_size;
@@ -1319,6 +1321,9 @@ private:
     /// Handles writes to the macro bind register.
     void ProcessMacroBind(u32 data);
 
+    /// Handles firmware blob 4
+    void ProcessFirmwareCall4();
+
     /// Handles a write to the CLEAR_BUFFERS register.
     void ProcessClearBuffers();
 
@@ -1431,6 +1436,7 @@ ASSERT_REG_POSITION(vertex_array[0], 0x700);
 ASSERT_REG_POSITION(independent_blend, 0x780);
 ASSERT_REG_POSITION(vertex_array_limit[0], 0x7C0);
 ASSERT_REG_POSITION(shader_config[0], 0x800);
+ASSERT_REG_POSITION(firmware, 0x8C0);
 ASSERT_REG_POSITION(const_buffer, 0x8E0);
 ASSERT_REG_POSITION(cb_bind[0], 0x904);
 ASSERT_REG_POSITION(tex_cb_index, 0x982);

src/video_core/engines/shader_bytecode.h

@@ -544,6 +544,35 @@ enum class VoteOperation : u64 {
     Eq = 2,  // allThreadsEqualNV
 };
 
+enum class ImageAtomicSize : u64 {
+    U32 = 0,
+    S32 = 1,
+    U64 = 2,
+    F32 = 3,
+    S64 = 5,
+    SD32 = 6,
+    SD64 = 7,
+};
+
+enum class ImageAtomicOperation : u64 {
+    Add = 0,
+    Min = 1,
+    Max = 2,
+    Inc = 3,
+    Dec = 4,
+    And = 5,
+    Or = 6,
+    Xor = 7,
+    Exch = 8,
+};
+
+enum class ShuffleOperation : u64 {
+    Idx = 0,  // shuffleNV
+    Up = 1,   // shuffleUpNV
+    Down = 2, // shuffleDownNV
+    Bfly = 3, // shuffleXorNV
+};
+
 union Instruction {
     Instruction& operator=(const Instruction& instr) {
         value = instr.value;
@@ -577,6 +606,15 @@ union Instruction {
         BitField<42, 1, u64> negate_value;
     } vote;
 
+    union {
+        BitField<30, 2, ShuffleOperation> operation;
+        BitField<48, 3, u64> pred48;
+        BitField<28, 1, u64> is_index_imm;
+        BitField<29, 1, u64> is_mask_imm;
+        BitField<20, 5, u64> index_imm;
+        BitField<34, 13, u64> mask_imm;
+    } shfl;
+
     union {
         BitField<8, 8, Register> gpr;
         BitField<20, 24, s64> offset;
@@ -674,6 +712,10 @@ union Instruction {
         BitField<48, 1, u64> is_signed;
     } shift;
 
+    union {
+        BitField<39, 1, u64> wrap;
+    } shr;
+
     union {
         BitField<39, 5, u64> shift_amount;
         BitField<48, 1, u64> negate_b;
@@ -1387,6 +1429,14 @@ union Instruction {
         }
     } sust;
 
+    union {
+        BitField<28, 1, u64> is_ba;
+        BitField<51, 3, ImageAtomicSize> size;
+        BitField<33, 3, ImageType> image_type;
+        BitField<29, 4, ImageAtomicOperation> operation;
+        BitField<49, 2, OutOfBoundsStore> out_of_bounds_store;
+    } suatom_d;
+
     union {
         BitField<20, 24, u64> target;
         BitField<5, 1, u64> constant_buffer;
@@ -1508,6 +1558,7 @@ public:
         BRK,
         DEPBAR,
         VOTE,
+        SHFL,
         BFE_C,
         BFE_R,
         BFE_IMM,
@@ -1539,6 +1590,7 @@ public:
         TMML_B, // Texture Mip Map Level
         TMML,   // Texture Mip Map Level
         SUST,   // Surface Store
+        SUATOM, // Surface Atomic Operation
         EXIT,
         NOP,
         IPA,
@@ -1798,6 +1850,7 @@ private:
             INST("111000110000----", Id::EXIT, Type::Flow, "EXIT"),
             INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"),
             INST("0101000011011---", Id::VOTE, Type::Warp, "VOTE"),
+            INST("1110111100010---", Id::SHFL, Type::Warp, "SHFL"),
             INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"),
             INST("1110111101001---", Id::LD_S, Type::Memory, "LD_S"),
             INST("1110111101000---", Id::LD_L, Type::Memory, "LD_L"),
@@ -1822,6 +1875,7 @@ private:
             INST("110111110110----", Id::TMML_B, Type::Texture, "TMML_B"),
             INST("1101111101011---", Id::TMML, Type::Texture, "TMML"),
             INST("11101011001-----", Id::SUST, Type::Image, "SUST"),
+            INST("1110101000------", Id::SUATOM, Type::Image, "SUATOM_D"),
             INST("0101000010110---", Id::NOP, Type::Trivial, "NOP"),
             INST("11100000--------", Id::IPA, Type::Trivial, "IPA"),
             INST("1111101111100---", Id::OUT_R, Type::Trivial, "OUT_R"),

src/video_core/macro_interpreter.cpp

@@ -131,9 +131,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
 
     // An instruction with the Exit flag will not actually
     // cause an exit if it's executed inside a delay slot.
-    // TODO(Blinkhawk): Reversed to always exit. The behavior explained above requires further
-    // testing on the MME code.
-    if (opcode.is_exit) {
+    if (opcode.is_exit && !is_delay_slot) {
         // Exit has a delay slot, execute the next instruction
         Step(offset, true);
         return false;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -489,9 +489,6 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
                 // Assume that a surface will be written to if it is used as a framebuffer, even if
                 // the shader doesn't actually write to it.
                 texture_cache.MarkColorBufferInUse(*single_color_target);
-                // Workaround for and issue in nvidia drivers
-                // https://devtalk.nvidia.com/default/topic/776591/opengl/gl_framebuffer_srgb-functions-incorrectly/
-                state.framebuffer_srgb.enabled |= color_surface->GetSurfaceParams().srgb_conversion;
             }
 
             fbkey.is_single_buffer = true;
@@ -512,11 +509,6 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
                     // Assume that a surface will be written to if it is used as a framebuffer, even
                     // if the shader doesn't actually write to it.
                     texture_cache.MarkColorBufferInUse(index);
-                    // Enable sRGB only for supported formats
-                    // Workaround for and issue in nvidia drivers
-                    // https://devtalk.nvidia.com/default/topic/776591/opengl/gl_framebuffer_srgb-functions-incorrectly/
-                    state.framebuffer_srgb.enabled |=
-                        color_surface->GetSurfaceParams().srgb_conversion;
                 }
 
                 fbkey.color_attachments[index] =
@@ -906,6 +898,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
     }
 
     screen_info.display_texture = surface->GetTexture();
+    screen_info.display_srgb = surface->GetSurfaceParams().srgb_conversion;
 
     return true;
 }

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -214,7 +214,8 @@ CachedProgram SpecializeShader(const std::string& code, const GLShader::ShaderEn
     std::string source = "#version 430 core\n"
                          "#extension GL_ARB_separate_shader_objects : enable\n"
                          "#extension GL_NV_gpu_shader5 : enable\n"
-                         "#extension GL_NV_shader_thread_group : enable\n";
+                         "#extension GL_NV_shader_thread_group : enable\n"
+                         "#extension GL_NV_shader_thread_shuffle : enable\n";
     if (entries.shader_viewport_layer_array) {
         source += "#extension GL_ARB_shader_viewport_layer_array : enable\n";
     }

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -325,6 +325,7 @@ public:
         DeclareRegisters();
         DeclarePredicates();
         DeclareLocalMemory();
+        DeclareSharedMemory();
         DeclareInternalFlags();
         DeclareInputAttributes();
         DeclareOutputAttributes();
@@ -499,6 +500,13 @@ private:
         code.AddNewLine();
     }
 
+    void DeclareSharedMemory() {
+        if (stage != ProgramType::Compute) {
+            return;
+        }
+        code.AddLine("shared uint {}[];", GetSharedMemory());
+    }
+
     void DeclareInternalFlags() {
         for (u32 flag = 0; flag < static_cast<u32>(InternalFlag::Amount); flag++) {
             const auto flag_code = static_cast<InternalFlag>(flag);
@@ -706,7 +714,7 @@ private:
     void DeclareImages() {
         const auto& images{ir.GetImages()};
         for (const auto& [offset, image] : images) {
-            const std::string image_type = [&]() {
+            const char* image_type = [&] {
                 switch (image.GetType()) {
                 case Tegra::Shader::ImageType::Texture1D:
                     return "image1D";
@@ -725,6 +733,23 @@ private:
                     return "image1D";
                 }
             }();
+
+            const auto [type_prefix, format] = [&]() -> std::pair<const char*, const char*> {
+                if (!image.IsSizeKnown()) {
+                    return {"", ""};
+                }
+                switch (image.GetSize()) {
+                case Tegra::Shader::ImageAtomicSize::U32:
+                    return {"u", "r32ui, "};
+                case Tegra::Shader::ImageAtomicSize::S32:
+                    return {"i", "r32i, "};
+                default:
+                    UNIMPLEMENTED_MSG("Unimplemented atomic size={}",
+                                      static_cast<u32>(image.GetSize()));
+                    return {"", ""};
+                }
+            }();
+
             std::string qualifier = "coherent volatile";
             if (image.IsRead() && !image.IsWritten()) {
                 qualifier += " readonly";
@@ -864,6 +889,12 @@ private:
                 Type::Uint};
         }
 
+        if (const auto smem = std::get_if<SmemNode>(&*node)) {
+            return {
+                fmt::format("{}[{} >> 2]", GetSharedMemory(), Visit(smem->GetAddress()).AsUint()),
+                Type::Uint};
+        }
+
         if (const auto internal_flag = std::get_if<InternalFlagNode>(&*node)) {
             return {GetInternalFlag(internal_flag->GetFlag()), Type::Bool};
         }
@@ -990,10 +1021,10 @@ private:
         return {std::move(temporary), value.GetType()};
     }
 
-    Expression GetOutputAttribute(const AbufNode* abuf) {
+    std::optional<Expression> GetOutputAttribute(const AbufNode* abuf) {
         switch (const auto attribute = abuf->GetIndex()) {
         case Attribute::Index::Position:
-            return {"gl_Position"s + GetSwizzle(abuf->GetElement()), Type::Float};
+            return {{"gl_Position"s + GetSwizzle(abuf->GetElement()), Type::Float}};
         case Attribute::Index::LayerViewportPointSize:
             switch (abuf->GetElement()) {
             case 0:
@@ -1003,25 +1034,25 @@ private:
                 if (IsVertexShader(stage) && !device.HasVertexViewportLayer()) {
                     return {};
                 }
-                return {"gl_Layer", Type::Int};
+                return {{"gl_Layer", Type::Int}};
             case 2:
                 if (IsVertexShader(stage) && !device.HasVertexViewportLayer()) {
                     return {};
                 }
-                return {"gl_ViewportIndex", Type::Int};
+                return {{"gl_ViewportIndex", Type::Int}};
             case 3:
                 UNIMPLEMENTED_MSG("Requires some state changes for gl_PointSize to work in shader");
-                return {"gl_PointSize", Type::Float};
+                return {{"gl_PointSize", Type::Float}};
             }
             return {};
         case Attribute::Index::ClipDistances0123:
-            return {fmt::format("gl_ClipDistance[{}]", abuf->GetElement()), Type::Float};
+            return {{fmt::format("gl_ClipDistance[{}]", abuf->GetElement()), Type::Float}};
         case Attribute::Index::ClipDistances4567:
-            return {fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4), Type::Float};
+            return {{fmt::format("gl_ClipDistance[{}]", abuf->GetElement() + 4), Type::Float}};
         default:
             if (IsGenericAttribute(attribute)) {
-                return {GetOutputAttribute(attribute) + GetSwizzle(abuf->GetElement()),
-                        Type::Float};
+                return {
+                    {GetOutputAttribute(attribute) + GetSwizzle(abuf->GetElement()), Type::Float}};
             }
             UNIMPLEMENTED_MSG("Unhandled output attribute: {}", static_cast<u32>(attribute));
             return {};
@@ -1180,6 +1211,74 @@ private:
         return expr;
     }
 
+    std::string BuildIntegerCoordinates(Operation operation) {
+        constexpr std::array constructors{"int(", "ivec2(", "ivec3(", "ivec4("};
+        const std::size_t coords_count{operation.GetOperandsCount()};
+        std::string expr = constructors.at(coords_count - 1);
+        for (std::size_t i = 0; i < coords_count; ++i) {
+            expr += VisitOperand(operation, i).AsInt();
+            if (i + 1 < coords_count) {
+                expr += ", ";
+            }
+        }
+        expr += ')';
+        return expr;
+    }
+
+    std::string BuildImageValues(Operation operation) {
+        const auto meta{std::get<MetaImage>(operation.GetMeta())};
+        const auto [constructors, type] = [&]() -> std::pair<std::array<const char*, 4>, Type> {
+            constexpr std::array float_constructors{"float", "vec2", "vec3", "vec4"};
+            if (!meta.image.IsSizeKnown()) {
+                return {float_constructors, Type::Float};
+            }
+            switch (meta.image.GetSize()) {
+            case Tegra::Shader::ImageAtomicSize::U32:
+                return {{"uint", "uvec2", "uvec3", "uvec4"}, Type::Uint};
+            case Tegra::Shader::ImageAtomicSize::S32:
+                return {{"int", "ivec2", "ivec3", "ivec4"}, Type::Uint};
+            default:
+                UNIMPLEMENTED_MSG("Unimplemented image size={}",
+                                  static_cast<u32>(meta.image.GetSize()));
+                return {float_constructors, Type::Float};
+            }
+        }();
+
+        const std::size_t values_count{meta.values.size()};
+        std::string expr = fmt::format("{}(", constructors.at(values_count - 1));
+        for (std::size_t i = 0; i < values_count; ++i) {
+            expr += Visit(meta.values.at(i)).As(type);
+            if (i + 1 < values_count) {
+                expr += ", ";
+            }
+        }
+        expr += ')';
+        return expr;
+    }
+
+    Expression AtomicImage(Operation operation, const char* opname) {
+        constexpr std::array constructors{"int(", "ivec2(", "ivec3(", "ivec4("};
+        const auto meta{std::get<MetaImage>(operation.GetMeta())};
+        ASSERT(meta.values.size() == 1);
+        ASSERT(meta.image.IsSizeKnown());
+
+        const auto type = [&]() {
+            switch (const auto size = meta.image.GetSize()) {
+            case Tegra::Shader::ImageAtomicSize::U32:
+                return Type::Uint;
+            case Tegra::Shader::ImageAtomicSize::S32:
+                return Type::Int;
+            default:
+                UNIMPLEMENTED_MSG("Unimplemented image size={}", static_cast<u32>(size));
+                return Type::Uint;
+            }
+        }();
+
+        return {fmt::format("{}({}, {}, {})", opname, GetImage(meta.image),
+                            BuildIntegerCoordinates(operation), Visit(meta.values[0]).As(type)),
+                type};
+    }
+
     Expression Assign(Operation operation) {
         const Node& dest = operation[0];
         const Node& src = operation[1];
@@ -1193,7 +1292,11 @@ private:
             target = {GetRegister(gpr->GetIndex()), Type::Float};
         } else if (const auto abuf = std::get_if<AbufNode>(&*dest)) {
             UNIMPLEMENTED_IF(abuf->IsPhysicalBuffer());
-            target = GetOutputAttribute(abuf);
+            auto output = GetOutputAttribute(abuf);
+            if (!output) {
+                return {};
+            }
+            target = std::move(*output);
         } else if (const auto lmem = std::get_if<LmemNode>(&*dest)) {
             if (stage == ProgramType::Compute) {
                 LOG_WARNING(Render_OpenGL, "Local memory is stubbed on compute shaders");
@@ -1201,6 +1304,11 @@ private:
             target = {
                 fmt::format("{}[{} >> 2]", GetLocalMemory(), Visit(lmem->GetAddress()).AsUint()),
                 Type::Uint};
+        } else if (const auto smem = std::get_if<SmemNode>(&*dest)) {
+            ASSERT(stage == ProgramType::Compute);
+            target = {
+                fmt::format("{}[{} >> 2]", GetSharedMemory(), Visit(smem->GetAddress()).AsUint()),
+                Type::Uint};
         } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) {
             const std::string real = Visit(gmem->GetRealAddress()).AsUint();
             const std::string base = Visit(gmem->GetBaseAddress()).AsUint();
@@ -1694,38 +1802,39 @@ private:
     }
 
     Expression ImageStore(Operation operation) {
-        constexpr std::array constructors{"int(", "ivec2(", "ivec3(", "ivec4("};
         const auto meta{std::get<MetaImage>(operation.GetMeta())};
-
-        std::string expr = "imageStore(";
-        expr += GetImage(meta.image);
-        expr += ", ";
-
-        const std::size_t coords_count{operation.GetOperandsCount()};
-        expr += constructors.at(coords_count - 1);
-        for (std::size_t i = 0; i < coords_count; ++i) {
-            expr += VisitOperand(operation, i).AsInt();
-            if (i + 1 < coords_count) {
-                expr += ", ";
-            }
-        }
-        expr += "), ";
-
-        const std::size_t values_count{meta.values.size()};
-        UNIMPLEMENTED_IF(values_count != 4);
-        expr += "vec4(";
-        for (std::size_t i = 0; i < values_count; ++i) {
-            expr += Visit(meta.values.at(i)).AsFloat();
-            if (i + 1 < values_count) {
-                expr += ", ";
-            }
-        }
-        expr += "));";
-
-        code.AddLine(expr);
+        code.AddLine("imageStore({}, {}, {});", GetImage(meta.image),
+                     BuildIntegerCoordinates(operation), BuildImageValues(operation));
         return {};
     }
 
+    Expression AtomicImageAdd(Operation operation) {
+        return AtomicImage(operation, "imageAtomicAdd");
+    }
+
+    Expression AtomicImageMin(Operation operation) {
+        return AtomicImage(operation, "imageAtomicMin");
+    }
+
+    Expression AtomicImageMax(Operation operation) {
+        return AtomicImage(operation, "imageAtomicMax");
+    }
+    Expression AtomicImageAnd(Operation operation) {
+        return AtomicImage(operation, "imageAtomicAnd");
+    }
+
+    Expression AtomicImageOr(Operation operation) {
+        return AtomicImage(operation, "imageAtomicOr");
+    }
+
+    Expression AtomicImageXor(Operation operation) {
+        return AtomicImage(operation, "imageAtomicXor");
+    }
+
+    Expression AtomicImageExchange(Operation operation) {
+        return AtomicImage(operation, "imageAtomicExchange");
+    }
+
     Expression Branch(Operation operation) {
         const auto target = std::get_if<ImmediateNode>(&*operation[0]);
         UNIMPLEMENTED_IF(!target);
@@ -1848,8 +1957,7 @@ private:
     Expression BallotThread(Operation operation) {
         const std::string value = VisitOperand(operation, 0).AsBool();
         if (!device.HasWarpIntrinsics()) {
-            LOG_ERROR(Render_OpenGL,
-                      "Nvidia warp intrinsics are not available and its required by a shader");
+            LOG_ERROR(Render_OpenGL, "Nvidia vote intrinsics are required by this shader");
             // Stub on non-Nvidia devices by simulating all threads voting the same as the active
             // one.
             return {fmt::format("({} ? 0xFFFFFFFFU : 0U)", value), Type::Uint};
@@ -1860,8 +1968,7 @@ private:
     Expression Vote(Operation operation, const char* func) {
         const std::string value = VisitOperand(operation, 0).AsBool();
         if (!device.HasWarpIntrinsics()) {
-            LOG_ERROR(Render_OpenGL,
-                      "Nvidia vote intrinsics are not available and its required by a shader");
+            LOG_ERROR(Render_OpenGL, "Nvidia vote intrinsics are required by this shader");
             // Stub with a warp size of one.
             return {value, Type::Bool};
         }
@@ -1878,15 +1985,54 @@ private:
 
     Expression VoteEqual(Operation operation) {
         if (!device.HasWarpIntrinsics()) {
-            LOG_ERROR(Render_OpenGL,
-                      "Nvidia vote intrinsics are not available and its required by a shader");
-            // We must return true here since a stub for a theoretical warp size of 1 will always
-            // return an equal result for all its votes.
+            LOG_ERROR(Render_OpenGL, "Nvidia vote intrinsics are required by this shader");
+            // We must return true here since a stub for a theoretical warp size of 1.
+            // This will always return an equal result across all votes.
             return {"true", Type::Bool};
         }
         return Vote(operation, "allThreadsEqualNV");
     }
 
+    template <const std::string_view& func>
+    Expression Shuffle(Operation operation) {
+        const std::string value = VisitOperand(operation, 0).AsFloat();
+        if (!device.HasWarpIntrinsics()) {
+            LOG_ERROR(Render_OpenGL, "Nvidia shuffle intrinsics are required by this shader");
+            // On a "single-thread" device we are either on the same thread or out of bounds. Both
+            // cases return the passed value.
+            return {value, Type::Float};
+        }
+
+        const std::string index = VisitOperand(operation, 1).AsUint();
+        const std::string width = VisitOperand(operation, 2).AsUint();
+        return {fmt::format("{}({}, {}, {})", func, value, index, width), Type::Float};
+    }
+
+    template <const std::string_view& func>
+    Expression InRangeShuffle(Operation operation) {
+        const std::string index = VisitOperand(operation, 0).AsUint();
+        const std::string width = VisitOperand(operation, 1).AsUint();
+        if (!device.HasWarpIntrinsics()) {
+            // On a "single-thread" device we are only in bounds when the requested index is 0.
+            return {fmt::format("({} == 0U)", index), Type::Bool};
+        }
+
+        const std::string in_range = code.GenerateTemporary();
+        code.AddLine("bool {};", in_range);
+        code.AddLine("{}(0U, {}, {}, {});", func, index, width, in_range);
+        return {in_range, Type::Bool};
+    }
+
+    struct Func final {
+        Func() = delete;
+        ~Func() = delete;
+
+        static constexpr std::string_view ShuffleIndexed = "shuffleNV";
+        static constexpr std::string_view ShuffleUp = "shuffleUpNV";
+        static constexpr std::string_view ShuffleDown = "shuffleDownNV";
+        static constexpr std::string_view ShuffleButterfly = "shuffleXorNV";
+    };
+
     static constexpr std::array operation_decompilers = {
         &GLSLDecompiler::Assign,
 
@@ -2019,6 +2165,13 @@ private:
         &GLSLDecompiler::TexelFetch,
 
         &GLSLDecompiler::ImageStore,
+        &GLSLDecompiler::AtomicImageAdd,
+        &GLSLDecompiler::AtomicImageMin,
+        &GLSLDecompiler::AtomicImageMax,
+        &GLSLDecompiler::AtomicImageAnd,
+        &GLSLDecompiler::AtomicImageOr,
+        &GLSLDecompiler::AtomicImageXor,
+        &GLSLDecompiler::AtomicImageExchange,
 
         &GLSLDecompiler::Branch,
         &GLSLDecompiler::BranchIndirect,
@@ -2042,6 +2195,16 @@ private:
         &GLSLDecompiler::VoteAll,
         &GLSLDecompiler::VoteAny,
         &GLSLDecompiler::VoteEqual,
+
+        &GLSLDecompiler::Shuffle<Func::ShuffleIndexed>,
+        &GLSLDecompiler::Shuffle<Func::ShuffleUp>,
+        &GLSLDecompiler::Shuffle<Func::ShuffleDown>,
+        &GLSLDecompiler::Shuffle<Func::ShuffleButterfly>,
+
+        &GLSLDecompiler::InRangeShuffle<Func::ShuffleIndexed>,
+        &GLSLDecompiler::InRangeShuffle<Func::ShuffleUp>,
+        &GLSLDecompiler::InRangeShuffle<Func::ShuffleDown>,
+        &GLSLDecompiler::InRangeShuffle<Func::ShuffleButterfly>,
     };
     static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
 
@@ -2082,6 +2245,10 @@ private:
         return "lmem_" + suffix;
     }
 
+    std::string GetSharedMemory() const {
+        return fmt::format("smem_{}", suffix);
+    }
+
     std::string GetInternalFlag(InternalFlag flag) const {
         constexpr std::array InternalFlagNames = {"zero_flag", "sign_flag", "carry_flag",
                                                   "overflow_flag"};

src/video_core/renderer_opengl/gl_shader_disk_cache.cpp

@@ -341,16 +341,22 @@ std::optional<ShaderDiskCacheDecompiled> ShaderDiskCacheOpenGL::LoadDecompiledEn
         u64 index{};
         u32 type{};
         u8 is_bindless{};
-        u8 is_read{};
         u8 is_written{};
+        u8 is_read{};
+        u8 is_size_known{};
+        u32 size{};
         if (!LoadObjectFromPrecompiled(offset) || !LoadObjectFromPrecompiled(index) ||
             !LoadObjectFromPrecompiled(type) || !LoadObjectFromPrecompiled(is_bindless) ||
-            !LoadObjectFromPrecompiled(is_read) || !LoadObjectFromPrecompiled(is_written)) {
+            !LoadObjectFromPrecompiled(is_written) || !LoadObjectFromPrecompiled(is_read) ||
+            !LoadObjectFromPrecompiled(is_size_known) || !LoadObjectFromPrecompiled(size)) {
             return {};
         }
-        entry.entries.images.emplace_back(static_cast<u64>(offset), static_cast<std::size_t>(index),
-                                          static_cast<Tegra::Shader::ImageType>(type),
-                                          is_bindless != 0, is_written != 0, is_read != 0);
+        entry.entries.images.emplace_back(
+            static_cast<std::size_t>(offset), static_cast<std::size_t>(index),
+            static_cast<Tegra::Shader::ImageType>(type), is_bindless != 0, is_written != 0,
+            is_read != 0,
+            is_size_known ? std::make_optional(static_cast<Tegra::Shader::ImageAtomicSize>(size))
+                          : std::nullopt);
     }
 
     u32 global_memory_count{};
@@ -429,12 +435,14 @@ bool ShaderDiskCacheOpenGL::SaveDecompiledFile(u64 unique_identifier, const std:
         return false;
     }
     for (const auto& image : entries.images) {
+        const u32 size = image.IsSizeKnown() ? static_cast<u32>(image.GetSize()) : 0U;
         if (!SaveObjectToPrecompiled(static_cast<u64>(image.GetOffset())) ||
             !SaveObjectToPrecompiled(static_cast<u64>(image.GetIndex())) ||
             !SaveObjectToPrecompiled(static_cast<u32>(image.GetType())) ||
             !SaveObjectToPrecompiled(static_cast<u8>(image.IsBindless() ? 1 : 0)) ||
+            !SaveObjectToPrecompiled(static_cast<u8>(image.IsWritten() ? 1 : 0)) ||
             !SaveObjectToPrecompiled(static_cast<u8>(image.IsRead() ? 1 : 0)) ||
-            !SaveObjectToPrecompiled(static_cast<u8>(image.IsWritten() ? 1 : 0))) {
+            !SaveObjectToPrecompiled(image.IsSizeKnown()) || !SaveObjectToPrecompiled(size)) {
             return false;
         }
     }

src/video_core/renderer_opengl/gl_state.cpp

@@ -16,7 +16,6 @@ namespace OpenGL {
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
 OpenGLState OpenGLState::cur_state;
-bool OpenGLState::s_rgb_used;
 
 namespace {
 
@@ -282,8 +281,6 @@ void OpenGLState::ApplySRgb() const {
         return;
     cur_state.framebuffer_srgb.enabled = framebuffer_srgb.enabled;
     if (framebuffer_srgb.enabled) {
-        // Track if sRGB is used
-        s_rgb_used = true;
         glEnable(GL_FRAMEBUFFER_SRGB);
     } else {
         glDisable(GL_FRAMEBUFFER_SRGB);

src/video_core/renderer_opengl/gl_state.h

@@ -175,14 +175,6 @@ public:
         return cur_state;
     }
 
-    static bool GetsRGBUsed() {
-        return s_rgb_used;
-    }
-
-    static void ClearsRGBUsed() {
-        s_rgb_used = false;
-    }
-
     void SetDefaultViewports();
     /// Apply this state as the current OpenGL state
     void Apply();
@@ -253,8 +245,6 @@ public:
 private:
     static OpenGLState cur_state;
 
-    // Workaround for sRGB problems caused by QT not supporting srgb output
-    static bool s_rgb_used;
     struct {
         bool blend_state;
         bool stencil_state;

src/video_core/renderer_opengl/maxwell_to_gl.h

@@ -145,7 +145,7 @@ inline GLenum TextureFilterMode(Tegra::Texture::TextureFilter filter_mode,
         case Tegra::Texture::TextureMipmapFilter::None:
             return GL_LINEAR;
         case Tegra::Texture::TextureMipmapFilter::Nearest:
-            return GL_NEAREST_MIPMAP_LINEAR;
+            return GL_LINEAR_MIPMAP_NEAREST;
         case Tegra::Texture::TextureMipmapFilter::Linear:
             return GL_LINEAR_MIPMAP_LINEAR;
         }
@@ -157,7 +157,7 @@ inline GLenum TextureFilterMode(Tegra::Texture::TextureFilter filter_mode,
         case Tegra::Texture::TextureMipmapFilter::Nearest:
             return GL_NEAREST_MIPMAP_NEAREST;
         case Tegra::Texture::TextureMipmapFilter::Linear:
-            return GL_LINEAR_MIPMAP_NEAREST;
+            return GL_NEAREST_MIPMAP_LINEAR;
         }
     }
     }

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -264,7 +264,6 @@ void RendererOpenGL::CreateRasterizer() {
     if (rasterizer) {
         return;
     }
-    OpenGLState::ClearsRGBUsed();
     rasterizer = std::make_unique<RasterizerOpenGL>(system, emu_window, screen_info);
 }
 
@@ -343,9 +342,7 @@ void RendererOpenGL::DrawScreenTriangles(const ScreenInfo& screen_info, float x,
     }};
 
     state.textures[0] = screen_info.display_texture;
-    // Workaround brigthness problems in SMO by enabling sRGB in the final output
-    // if it has been used in the frame. Needed because of this bug in QT: QTBUG-50987
-    state.framebuffer_srgb.enabled = OpenGLState::GetsRGBUsed();
+    state.framebuffer_srgb.enabled = screen_info.display_srgb;
     state.AllDirty();
     state.Apply();
     glNamedBufferSubData(vertex_buffer.handle, 0, sizeof(vertices), vertices.data());
@@ -355,8 +352,6 @@ void RendererOpenGL::DrawScreenTriangles(const ScreenInfo& screen_info, float x,
     state.textures[0] = 0;
     state.AllDirty();
     state.Apply();
-    // Clear sRGB state for the next frame
-    OpenGLState::ClearsRGBUsed();
 }
 
 /**
@@ -406,8 +401,8 @@ void RendererOpenGL::CaptureScreenshot() {
     GLuint renderbuffer;
     glGenRenderbuffers(1, &renderbuffer);
     glBindRenderbuffer(GL_RENDERBUFFER, renderbuffer);
-    glRenderbufferStorage(GL_RENDERBUFFER, state.GetsRGBUsed() ? GL_SRGB8 : GL_RGB8, layout.width,
-                          layout.height);
+    glRenderbufferStorage(GL_RENDERBUFFER, screen_info.display_srgb ? GL_SRGB8 : GL_RGB8,
+                          layout.width, layout.height);
     glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_RENDERBUFFER, renderbuffer);
 
     DrawScreen(layout);

src/video_core/renderer_opengl/renderer_opengl.h

@@ -38,7 +38,8 @@ struct TextureInfo {
 
 /// Structure used for storing information about the display target for the Switch screen
 struct ScreenInfo {
-    GLuint display_texture;
+    GLuint display_texture{};
+    bool display_srgb{};
     const Common::Rectangle<float> display_texcoords{0.0f, 0.0f, 1.0f, 1.0f};
     TextureInfo texture;
 };

src/video_core/renderer_vulkan/vk_device.cpp

@@ -2,9 +2,10 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <map>
+#include <bitset>
 #include <optional>
 #include <set>
+#include <string_view>
 #include <vector>
 #include "common/assert.h"
 #include "video_core/renderer_vulkan/declarations.h"
@@ -12,13 +13,32 @@
 
 namespace Vulkan {
 
+namespace {
+
+template <typename T>
+void SetNext(void**& next, T& data) {
+    *next = &data;
+    next = &data.pNext;
+}
+
+template <typename T>
+T GetFeatures(vk::PhysicalDevice physical, vk::DispatchLoaderDynamic dldi) {
+    vk::PhysicalDeviceFeatures2 features;
+    T extension_features;
+    features.pNext = &extension_features;
+    physical.getFeatures2(&features, dldi);
+    return extension_features;
+}
+
+} // Anonymous namespace
+
 namespace Alternatives {
 
-constexpr std::array<vk::Format, 3> Depth24UnormS8Uint = {
-    vk::Format::eD32SfloatS8Uint, vk::Format::eD16UnormS8Uint, {}};
-constexpr std::array<vk::Format, 3> Depth16UnormS8Uint = {
-    vk::Format::eD24UnormS8Uint, vk::Format::eD32SfloatS8Uint, {}};
-constexpr std::array<vk::Format, 2> Astc = {vk::Format::eA8B8G8R8UnormPack32, {}};
+constexpr std::array Depth24UnormS8Uint = {vk::Format::eD32SfloatS8Uint,
+                                           vk::Format::eD16UnormS8Uint, vk::Format{}};
+constexpr std::array Depth16UnormS8Uint = {vk::Format::eD24UnormS8Uint,
+                                           vk::Format::eD32SfloatS8Uint, vk::Format{}};
+constexpr std::array Astc = {vk::Format::eA8B8G8R8UnormPack32, vk::Format{}};
 
 } // namespace Alternatives
 
@@ -58,16 +78,53 @@ VKDevice::VKDevice(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice phy
 VKDevice::~VKDevice() = default;
 
 bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instance) {
-    vk::PhysicalDeviceFeatures device_features;
-    device_features.vertexPipelineStoresAndAtomics = true;
-    device_features.independentBlend = true;
-    device_features.textureCompressionASTC_LDR = is_optimal_astc_supported;
-
     const auto queue_cis = GetDeviceQueueCreateInfos();
-    const std::vector<const char*> extensions = LoadExtensions(dldi);
-    const vk::DeviceCreateInfo device_ci({}, static_cast<u32>(queue_cis.size()), queue_cis.data(),
-                                         0, nullptr, static_cast<u32>(extensions.size()),
-                                         extensions.data(), &device_features);
+    const std::vector extensions = LoadExtensions(dldi);
+
+    vk::PhysicalDeviceFeatures2 features2;
+    void** next = &features2.pNext;
+    auto& features = features2.features;
+    features.vertexPipelineStoresAndAtomics = true;
+    features.independentBlend = true;
+    features.depthClamp = true;
+    features.samplerAnisotropy = true;
+    features.largePoints = true;
+    features.textureCompressionASTC_LDR = is_optimal_astc_supported;
+
+    vk::PhysicalDeviceVertexAttributeDivisorFeaturesEXT vertex_divisor;
+    vertex_divisor.vertexAttributeInstanceRateDivisor = true;
+    vertex_divisor.vertexAttributeInstanceRateZeroDivisor = true;
+    SetNext(next, vertex_divisor);
+
+    vk::PhysicalDeviceFloat16Int8FeaturesKHR float16_int8;
+    if (is_float16_supported) {
+        float16_int8.shaderFloat16 = true;
+        SetNext(next, float16_int8);
+    } else {
+        LOG_INFO(Render_Vulkan, "Device doesn't support float16 natively");
+    }
+
+    vk::PhysicalDeviceUniformBufferStandardLayoutFeaturesKHR std430_layout;
+    if (khr_uniform_buffer_standard_layout) {
+        std430_layout.uniformBufferStandardLayout = true;
+        SetNext(next, std430_layout);
+    } else {
+        LOG_INFO(Render_Vulkan, "Device doesn't support packed UBOs");
+    }
+
+    vk::PhysicalDeviceIndexTypeUint8FeaturesEXT index_type_uint8;
+    if (ext_index_type_uint8) {
+        index_type_uint8.indexTypeUint8 = true;
+        SetNext(next, index_type_uint8);
+    } else {
+        LOG_INFO(Render_Vulkan, "Device doesn't support uint8 indexes");
+    }
+
+    vk::DeviceCreateInfo device_ci({}, static_cast<u32>(queue_cis.size()), queue_cis.data(), 0,
+                                   nullptr, static_cast<u32>(extensions.size()), extensions.data(),
+                                   nullptr);
+    device_ci.pNext = &features2;
+
     vk::Device dummy_logical;
     if (physical.createDevice(&device_ci, nullptr, &dummy_logical, dldi) != vk::Result::eSuccess) {
         LOG_CRITICAL(Render_Vulkan, "Logical device failed to be created!");
@@ -78,6 +135,17 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan
     logical = UniqueDevice(
         dummy_logical, vk::ObjectDestroy<vk::NoParent, vk::DispatchLoaderDynamic>(nullptr, dld));
 
+    if (khr_driver_properties) {
+        vk::PhysicalDeviceDriverPropertiesKHR driver;
+        vk::PhysicalDeviceProperties2 properties;
+        properties.pNext = &driver;
+        physical.getProperties2(&properties, dld);
+        driver_id = driver.driverID;
+        LOG_INFO(Render_Vulkan, "Driver: {} {}", driver.driverName, driver.driverInfo);
+    } else {
+        LOG_INFO(Render_Vulkan, "Driver: Unknown");
+    }
+
     graphics_queue = logical->getQueue(graphics_family, 0, dld);
     present_queue = logical->getQueue(present_family, 0, dld);
     return true;
@@ -92,20 +160,19 @@ vk::Format VKDevice::GetSupportedFormat(vk::Format wanted_format,
     // The wanted format is not supported by hardware, search for alternatives
     const vk::Format* alternatives = GetFormatAlternatives(wanted_format);
     if (alternatives == nullptr) {
-        LOG_CRITICAL(Render_Vulkan,
-                     "Format={} with usage={} and type={} has no defined alternatives and host "
-                     "hardware does not support it",
-                     vk::to_string(wanted_format), vk::to_string(wanted_usage),
-                     static_cast<u32>(format_type));
-        UNREACHABLE();
+        UNREACHABLE_MSG("Format={} with usage={} and type={} has no defined alternatives and host "
+                        "hardware does not support it",
+                        vk::to_string(wanted_format), vk::to_string(wanted_usage),
+                        static_cast<u32>(format_type));
         return wanted_format;
     }
 
     std::size_t i = 0;
     for (vk::Format alternative = alternatives[0]; alternative != vk::Format{};
          alternative = alternatives[++i]) {
-        if (!IsFormatSupported(alternative, wanted_usage, format_type))
+        if (!IsFormatSupported(alternative, wanted_usage, format_type)) {
             continue;
+        }
         LOG_WARNING(Render_Vulkan,
                     "Emulating format={} with alternative format={} with usage={} and type={}",
                     static_cast<u32>(wanted_format), static_cast<u32>(alternative),
@@ -114,12 +181,10 @@ vk::Format VKDevice::GetSupportedFormat(vk::Format wanted_format,
     }
 
     // No alternatives found, panic
-    LOG_CRITICAL(Render_Vulkan,
-                 "Format={} with usage={} and type={} is not supported by the host hardware and "
-                 "doesn't support any of the alternatives",
-                 static_cast<u32>(wanted_format), static_cast<u32>(wanted_usage),
-                 static_cast<u32>(format_type));
-    UNREACHABLE();
+    UNREACHABLE_MSG("Format={} with usage={} and type={} is not supported by the host hardware and "
+                    "doesn't support any of the alternatives",
+                    static_cast<u32>(wanted_format), static_cast<u32>(wanted_usage),
+                    static_cast<u32>(format_type));
     return wanted_format;
 }
 
@@ -132,7 +197,7 @@ bool VKDevice::IsOptimalAstcSupported(const vk::PhysicalDeviceFeatures& features
         vk::FormatFeatureFlagBits::eSampledImage | vk::FormatFeatureFlagBits::eBlitSrc |
         vk::FormatFeatureFlagBits::eBlitDst | vk::FormatFeatureFlagBits::eTransferSrc |
         vk::FormatFeatureFlagBits::eTransferDst};
-    constexpr std::array<vk::Format, 9> astc_formats = {
+    constexpr std::array astc_formats = {
         vk::Format::eAstc4x4UnormBlock, vk::Format::eAstc4x4SrgbBlock,
         vk::Format::eAstc8x8SrgbBlock,  vk::Format::eAstc8x6SrgbBlock,
         vk::Format::eAstc5x4SrgbBlock,  vk::Format::eAstc5x5UnormBlock,
@@ -151,76 +216,120 @@ bool VKDevice::IsFormatSupported(vk::Format wanted_format, vk::FormatFeatureFlag
                                  FormatType format_type) const {
     const auto it = format_properties.find(wanted_format);
     if (it == format_properties.end()) {
-        LOG_CRITICAL(Render_Vulkan, "Unimplemented format query={}", vk::to_string(wanted_format));
-        UNREACHABLE();
+        UNIMPLEMENTED_MSG("Unimplemented format query={}", vk::to_string(wanted_format));
         return true;
     }
-    const vk::FormatFeatureFlags supported_usage = GetFormatFeatures(it->second, format_type);
+    const auto supported_usage = GetFormatFeatures(it->second, format_type);
     return (supported_usage & wanted_usage) == wanted_usage;
 }
 
 bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical,
                           vk::SurfaceKHR surface) {
-    bool has_swapchain{};
+    LOG_INFO(Render_Vulkan, "{}", physical.getProperties(dldi).deviceName);
+    bool is_suitable = true;
+
+    constexpr std::array required_extensions = {VK_KHR_SWAPCHAIN_EXTENSION_NAME,
+                                                VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME};
+    std::bitset<required_extensions.size()> available_extensions{};
+
     for (const auto& prop : physical.enumerateDeviceExtensionProperties(nullptr, dldi)) {
-        has_swapchain |= prop.extensionName == std::string(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
+        for (std::size_t i = 0; i < required_extensions.size(); ++i) {
+            if (available_extensions[i]) {
+                continue;
+            }
+            available_extensions[i] =
+                required_extensions[i] == std::string_view{prop.extensionName};
+        }
     }
-    if (!has_swapchain) {
-        // The device doesn't support creating swapchains.
-        return false;
+    if (!available_extensions.all()) {
+        for (std::size_t i = 0; i < required_extensions.size(); ++i) {
+            if (available_extensions[i]) {
+                continue;
+            }
+            LOG_INFO(Render_Vulkan, "Missing required extension: {}", required_extensions[i]);
+            is_suitable = false;
+        }
     }
 
     bool has_graphics{}, has_present{};
     const auto queue_family_properties = physical.getQueueFamilyProperties(dldi);
     for (u32 i = 0; i < static_cast<u32>(queue_family_properties.size()); ++i) {
         const auto& family = queue_family_properties[i];
-        if (family.queueCount == 0)
+        if (family.queueCount == 0) {
             continue;
-
+        }
         has_graphics |=
             (family.queueFlags & vk::QueueFlagBits::eGraphics) != static_cast<vk::QueueFlagBits>(0);
         has_present |= physical.getSurfaceSupportKHR(i, surface, dldi) != 0;
     }
     if (!has_graphics || !has_present) {
-        // The device doesn't have a graphics and present queue.
-        return false;
+        LOG_INFO(Render_Vulkan, "Device lacks a graphics and present queue");
+        is_suitable = false;
     }
 
     // TODO(Rodrigo): Check if the device matches all requeriments.
     const auto properties{physical.getProperties(dldi)};
-    const auto limits{properties.limits};
-    if (limits.maxUniformBufferRange < 65536) {
-        return false;
+    const auto& limits{properties.limits};
+
+    constexpr u32 required_ubo_size = 65536;
+    if (limits.maxUniformBufferRange < required_ubo_size) {
+        LOG_INFO(Render_Vulkan, "Device UBO size {} is too small, {} is required)",
+                 limits.maxUniformBufferRange, required_ubo_size);
+        is_suitable = false;
     }
 
-    const vk::PhysicalDeviceFeatures features{physical.getFeatures(dldi)};
-    if (!features.vertexPipelineStoresAndAtomics || !features.independentBlend) {
-        return false;
+    const auto features{physical.getFeatures(dldi)};
+    const std::array feature_report = {
+        std::make_pair(features.vertexPipelineStoresAndAtomics, "vertexPipelineStoresAndAtomics"),
+        std::make_pair(features.independentBlend, "independentBlend"),
+        std::make_pair(features.depthClamp, "depthClamp"),
+        std::make_pair(features.samplerAnisotropy, "samplerAnisotropy"),
+        std::make_pair(features.largePoints, "largePoints"),
+    };
+    for (const auto& [supported, name] : feature_report) {
+        if (supported) {
+            continue;
+        }
+        LOG_INFO(Render_Vulkan, "Missing required feature: {}", name);
+        is_suitable = false;
     }
 
-    // Device is suitable.
-    return true;
+    return is_suitable;
 }
 
 std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynamic& dldi) {
     std::vector<const char*> extensions;
-    extensions.reserve(2);
+    extensions.reserve(7);
     extensions.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
+    extensions.push_back(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME);
 
     const auto Test = [&](const vk::ExtensionProperties& extension,
                           std::optional<std::reference_wrapper<bool>> status, const char* name,
-                          u32 revision) {
-        if (extension.extensionName != std::string(name)) {
+                          bool push) {
+        if (extension.extensionName != std::string_view(name)) {
             return;
         }
-        extensions.push_back(name);
+        if (push) {
+            extensions.push_back(name);
+        }
         if (status) {
             status->get() = true;
         }
     };
 
+    bool khr_shader_float16_int8{};
     for (const auto& extension : physical.enumerateDeviceExtensionProperties(nullptr, dldi)) {
-        Test(extension, ext_scalar_block_layout, VK_EXT_SCALAR_BLOCK_LAYOUT_EXTENSION_NAME, 1);
+        Test(extension, khr_uniform_buffer_standard_layout,
+             VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, true);
+        Test(extension, ext_index_type_uint8, VK_EXT_INDEX_TYPE_UINT8_EXTENSION_NAME, true);
+        Test(extension, khr_driver_properties, VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME, true);
+        Test(extension, khr_shader_float16_int8, VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME, false);
+    }
+
+    if (khr_shader_float16_int8) {
+        is_float16_supported =
+            GetFeatures<vk::PhysicalDeviceFloat16Int8FeaturesKHR>(physical, dldi).shaderFloat16;
+        extensions.push_back(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME);
     }
 
     return extensions;
@@ -250,9 +359,10 @@ void VKDevice::SetupFamilies(const vk::DispatchLoaderDynamic& dldi, vk::SurfaceK
 }
 
 void VKDevice::SetupProperties(const vk::DispatchLoaderDynamic& dldi) {
-    const vk::PhysicalDeviceProperties props = physical.getProperties(dldi);
+    const auto props = physical.getProperties(dldi);
     device_type = props.deviceType;
     uniform_buffer_alignment = static_cast<u64>(props.limits.minUniformBufferOffsetAlignment);
+    storage_buffer_alignment = static_cast<u64>(props.limits.minStorageBufferOffsetAlignment);
     max_storage_buffer_range = static_cast<u64>(props.limits.maxStorageBufferRange);
 }
 
@@ -273,42 +383,53 @@ std::vector<vk::DeviceQueueCreateInfo> VKDevice::GetDeviceQueueCreateInfos() con
     return queue_cis;
 }
 
-std::map<vk::Format, vk::FormatProperties> VKDevice::GetFormatProperties(
+std::unordered_map<vk::Format, vk::FormatProperties> VKDevice::GetFormatProperties(
     const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical) {
-    static constexpr std::array formats{vk::Format::eA8B8G8R8UnormPack32,
-                                        vk::Format::eB5G6R5UnormPack16,
-                                        vk::Format::eA2B10G10R10UnormPack32,
-                                        vk::Format::eR32G32B32A32Sfloat,
-                                        vk::Format::eR16G16Unorm,
-                                        vk::Format::eR16G16Snorm,
-                                        vk::Format::eR8G8B8A8Srgb,
-                                        vk::Format::eR8Unorm,
-                                        vk::Format::eB10G11R11UfloatPack32,
-                                        vk::Format::eR32Sfloat,
-                                        vk::Format::eR16Sfloat,
-                                        vk::Format::eR16G16B16A16Sfloat,
-                                        vk::Format::eD32Sfloat,
-                                        vk::Format::eD16Unorm,
-                                        vk::Format::eD16UnormS8Uint,
-                                        vk::Format::eD24UnormS8Uint,
-                                        vk::Format::eD32SfloatS8Uint,
-                                        vk::Format::eBc1RgbaUnormBlock,
-                                        vk::Format::eBc2UnormBlock,
-                                        vk::Format::eBc3UnormBlock,
-                                        vk::Format::eBc4UnormBlock,
-                                        vk::Format::eBc5UnormBlock,
-                                        vk::Format::eBc5SnormBlock,
-                                        vk::Format::eBc7UnormBlock,
-                                        vk::Format::eAstc4x4UnormBlock,
-                                        vk::Format::eAstc4x4SrgbBlock,
-                                        vk::Format::eAstc8x8SrgbBlock,
-                                        vk::Format::eAstc8x6SrgbBlock,
-                                        vk::Format::eAstc5x4SrgbBlock,
-                                        vk::Format::eAstc5x5UnormBlock,
-                                        vk::Format::eAstc5x5SrgbBlock,
-                                        vk::Format::eAstc10x8UnormBlock,
-                                        vk::Format::eAstc10x8SrgbBlock};
-    std::map<vk::Format, vk::FormatProperties> format_properties;
+    constexpr std::array formats{vk::Format::eA8B8G8R8UnormPack32,
+                                 vk::Format::eA8B8G8R8SnormPack32,
+                                 vk::Format::eA8B8G8R8SrgbPack32,
+                                 vk::Format::eB5G6R5UnormPack16,
+                                 vk::Format::eA2B10G10R10UnormPack32,
+                                 vk::Format::eR32G32B32A32Sfloat,
+                                 vk::Format::eR16G16B16A16Uint,
+                                 vk::Format::eR16G16Unorm,
+                                 vk::Format::eR16G16Snorm,
+                                 vk::Format::eR16G16Sfloat,
+                                 vk::Format::eR16Unorm,
+                                 vk::Format::eR8G8B8A8Srgb,
+                                 vk::Format::eR8G8Unorm,
+                                 vk::Format::eR8G8Snorm,
+                                 vk::Format::eR8Unorm,
+                                 vk::Format::eB10G11R11UfloatPack32,
+                                 vk::Format::eR32Sfloat,
+                                 vk::Format::eR16Sfloat,
+                                 vk::Format::eR16G16B16A16Sfloat,
+                                 vk::Format::eB8G8R8A8Unorm,
+                                 vk::Format::eD32Sfloat,
+                                 vk::Format::eD16Unorm,
+                                 vk::Format::eD16UnormS8Uint,
+                                 vk::Format::eD24UnormS8Uint,
+                                 vk::Format::eD32SfloatS8Uint,
+                                 vk::Format::eBc1RgbaUnormBlock,
+                                 vk::Format::eBc2UnormBlock,
+                                 vk::Format::eBc3UnormBlock,
+                                 vk::Format::eBc4UnormBlock,
+                                 vk::Format::eBc5UnormBlock,
+                                 vk::Format::eBc5SnormBlock,
+                                 vk::Format::eBc7UnormBlock,
+                                 vk::Format::eBc1RgbaSrgbBlock,
+                                 vk::Format::eBc3SrgbBlock,
+                                 vk::Format::eBc7SrgbBlock,
+                                 vk::Format::eAstc4x4UnormBlock,
+                                 vk::Format::eAstc4x4SrgbBlock,
+                                 vk::Format::eAstc8x8SrgbBlock,
+                                 vk::Format::eAstc8x6SrgbBlock,
+                                 vk::Format::eAstc5x4SrgbBlock,
+                                 vk::Format::eAstc5x5UnormBlock,
+                                 vk::Format::eAstc5x5SrgbBlock,
+                                 vk::Format::eAstc10x8UnormBlock,
+                                 vk::Format::eAstc10x8SrgbBlock};
+    std::unordered_map<vk::Format, vk::FormatProperties> format_properties;
     for (const auto format : formats) {
         format_properties.emplace(format, physical.getFormatProperties(format, dldi));
     }

src/video_core/renderer_vulkan/vk_device.h

@@ -4,7 +4,7 @@
 
 #pragma once
 
-#include <map>
+#include <unordered_map>
 #include <vector>
 #include "common/common_types.h"
 #include "video_core/renderer_vulkan/declarations.h"
@@ -69,16 +69,26 @@ public:
         return present_family;
     }
 
-    /// Returns if the device is integrated with the host CPU.
+    /// Returns true if the device is integrated with the host CPU.
     bool IsIntegrated() const {
         return device_type == vk::PhysicalDeviceType::eIntegratedGpu;
     }
 
+    /// Returns the driver ID.
+    vk::DriverIdKHR GetDriverID() const {
+        return driver_id;
+    }
+
     /// Returns uniform buffer alignment requeriment.
     u64 GetUniformBufferAlignment() const {
         return uniform_buffer_alignment;
     }
 
+    /// Returns storage alignment requeriment.
+    u64 GetStorageBufferAlignment() const {
+        return storage_buffer_alignment;
+    }
+
     /// Returns the maximum range for storage buffers.
     u64 GetMaxStorageBufferRange() const {
         return max_storage_buffer_range;
@@ -89,9 +99,19 @@ public:
         return is_optimal_astc_supported;
     }
 
+    /// Returns true if the device supports float16 natively
+    bool IsFloat16Supported() const {
+        return is_float16_supported;
+    }
+
     /// Returns true if the device supports VK_EXT_scalar_block_layout.
-    bool IsExtScalarBlockLayoutSupported() const {
-        return ext_scalar_block_layout;
+    bool IsKhrUniformBufferStandardLayoutSupported() const {
+        return khr_uniform_buffer_standard_layout;
+    }
+
+    /// Returns true if the device supports VK_EXT_index_type_uint8.
+    bool IsExtIndexTypeUint8Supported() const {
+        return ext_index_type_uint8;
     }
 
     /// Checks if the physical device is suitable.
@@ -123,22 +143,28 @@ private:
                            FormatType format_type) const;
 
     /// Returns the device properties for Vulkan formats.
-    static std::map<vk::Format, vk::FormatProperties> GetFormatProperties(
+    static std::unordered_map<vk::Format, vk::FormatProperties> GetFormatProperties(
         const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical);
 
-    const vk::PhysicalDevice physical;  ///< Physical device.
-    vk::DispatchLoaderDynamic dld;      ///< Device function pointers.
-    UniqueDevice logical;               ///< Logical device.
-    vk::Queue graphics_queue;           ///< Main graphics queue.
-    vk::Queue present_queue;            ///< Main present queue.
-    u32 graphics_family{};              ///< Main graphics queue family index.
-    u32 present_family{};               ///< Main present queue family index.
-    vk::PhysicalDeviceType device_type; ///< Physical device type.
-    u64 uniform_buffer_alignment{};     ///< Uniform buffer alignment requeriment.
-    u64 max_storage_buffer_range{};     ///< Max storage buffer size.
-    bool is_optimal_astc_supported{};   ///< Support for native ASTC.
-    bool ext_scalar_block_layout{};     ///< Support for VK_EXT_scalar_block_layout.
-    std::map<vk::Format, vk::FormatProperties> format_properties; ///< Format properties dictionary.
+    const vk::PhysicalDevice physical;         ///< Physical device.
+    vk::DispatchLoaderDynamic dld;             ///< Device function pointers.
+    UniqueDevice logical;                      ///< Logical device.
+    vk::Queue graphics_queue;                  ///< Main graphics queue.
+    vk::Queue present_queue;                   ///< Main present queue.
+    u32 graphics_family{};                     ///< Main graphics queue family index.
+    u32 present_family{};                      ///< Main present queue family index.
+    vk::PhysicalDeviceType device_type;        ///< Physical device type.
+    vk::DriverIdKHR driver_id{};               ///< Driver ID.
+    u64 uniform_buffer_alignment{};            ///< Uniform buffer alignment requeriment.
+    u64 storage_buffer_alignment{};            ///< Storage buffer alignment requeriment.
+    u64 max_storage_buffer_range{};            ///< Max storage buffer size.
+    bool is_optimal_astc_supported{};          ///< Support for native ASTC.
+    bool is_float16_supported{};               ///< Support for float16 arithmetics.
+    bool khr_uniform_buffer_standard_layout{}; ///< Support for std430 on UBOs.
+    bool ext_index_type_uint8{};               ///< Support for VK_EXT_index_type_uint8.
+    bool khr_driver_properties{};              ///< Support for VK_KHR_driver_properties.
+    std::unordered_map<vk::Format, vk::FormatProperties>
+        format_properties; ///< Format properties dictionary.
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -370,8 +370,8 @@ private:
         u32 binding = const_buffers_base_binding;
         for (const auto& entry : ir.GetConstantBuffers()) {
             const auto [index, size] = entry;
-            const Id type =
-                device.IsExtScalarBlockLayoutSupported() ? t_cbuf_scalar_ubo : t_cbuf_std140_ubo;
+            const Id type = device.IsKhrUniformBufferStandardLayoutSupported() ? t_cbuf_scalar_ubo
+                                                                               : t_cbuf_std140_ubo;
             const Id id = OpVariable(type, spv::StorageClass::Uniform);
             AddGlobalVariable(Name(id, fmt::format("cbuf_{}", index)));
 
@@ -565,7 +565,7 @@ private:
             const Id buffer_id = constant_buffers.at(cbuf->GetIndex());
 
             Id pointer{};
-            if (device.IsExtScalarBlockLayoutSupported()) {
+            if (device.IsKhrUniformBufferStandardLayoutSupported()) {
                 const Id buffer_offset = Emit(OpShiftRightLogical(
                     t_uint, BitcastTo<Type::Uint>(Visit(offset)), Constant(t_uint, 2u)));
                 pointer = Emit(
@@ -944,6 +944,41 @@ private:
         return {};
     }
 
+    Id AtomicImageAdd(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id AtomicImageMin(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id AtomicImageMax(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id AtomicImageAnd(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id AtomicImageOr(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id AtomicImageXor(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id AtomicImageExchange(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
     Id Branch(Operation operation) {
         const auto target = std::get_if<ImmediateNode>(&*operation[0]);
         UNIMPLEMENTED_IF(!target);
@@ -1092,6 +1127,46 @@ private:
         return {};
     }
 
+    Id ShuffleIndexed(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id ShuffleUp(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id ShuffleDown(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id ShuffleButterfly(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id InRangeShuffleIndexed(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id InRangeShuffleUp(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id InRangeShuffleDown(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
+    Id InRangeShuffleButterfly(Operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
     Id DeclareBuiltIn(spv::BuiltIn builtin, spv::StorageClass storage, Id type,
                       const std::string& name) {
         const Id id = OpVariable(type, storage);
@@ -1366,6 +1441,13 @@ private:
         &SPIRVDecompiler::TexelFetch,
 
         &SPIRVDecompiler::ImageStore,
+        &SPIRVDecompiler::AtomicImageAdd,
+        &SPIRVDecompiler::AtomicImageMin,
+        &SPIRVDecompiler::AtomicImageMax,
+        &SPIRVDecompiler::AtomicImageAnd,
+        &SPIRVDecompiler::AtomicImageOr,
+        &SPIRVDecompiler::AtomicImageXor,
+        &SPIRVDecompiler::AtomicImageExchange,
 
         &SPIRVDecompiler::Branch,
         &SPIRVDecompiler::BranchIndirect,
@@ -1389,6 +1471,16 @@ private:
         &SPIRVDecompiler::VoteAll,
         &SPIRVDecompiler::VoteAny,
         &SPIRVDecompiler::VoteEqual,
+
+        &SPIRVDecompiler::ShuffleIndexed,
+        &SPIRVDecompiler::ShuffleUp,
+        &SPIRVDecompiler::ShuffleDown,
+        &SPIRVDecompiler::ShuffleButterfly,
+
+        &SPIRVDecompiler::InRangeShuffleIndexed,
+        &SPIRVDecompiler::InRangeShuffleUp,
+        &SPIRVDecompiler::InRangeShuffleDown,
+        &SPIRVDecompiler::InRangeShuffleButterfly,
     };
     static_assert(operation_decompilers.size() == static_cast<std::size_t>(OperationCode::Amount));
 

src/video_core/shader/decode/image.cpp

@@ -44,7 +44,6 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
     switch (opcode->get().GetId()) {
     case OpCode::Id::SUST: {
         UNIMPLEMENTED_IF(instr.sust.mode != Tegra::Shader::SurfaceDataMode::P);
-        UNIMPLEMENTED_IF(instr.sust.image_type == Tegra::Shader::ImageType::TextureBuffer);
         UNIMPLEMENTED_IF(instr.sust.out_of_bounds_store != Tegra::Shader::OutOfBoundsStore::Ignore);
         UNIMPLEMENTED_IF(instr.sust.component_mask_selector != 0xf); // Ensure we have an RGBA store
 
@@ -66,8 +65,46 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
         image.MarkWrite();
 
         MetaImage meta{image, values};
-        const Node store{Operation(OperationCode::ImageStore, meta, std::move(coords))};
-        bb.push_back(store);
+        bb.push_back(Operation(OperationCode::ImageStore, meta, std::move(coords)));
+        break;
+    }
+    case OpCode::Id::SUATOM: {
+        UNIMPLEMENTED_IF(instr.suatom_d.is_ba != 0);
+
+        Node value = GetRegister(instr.gpr0);
+
+        std::vector<Node> coords;
+        const std::size_t num_coords{GetImageTypeNumCoordinates(instr.sust.image_type)};
+        for (std::size_t i = 0; i < num_coords; ++i) {
+            coords.push_back(GetRegister(instr.gpr8.Value() + i));
+        }
+
+        const OperationCode operation_code = [instr] {
+            switch (instr.suatom_d.operation) {
+            case Tegra::Shader::ImageAtomicOperation::Add:
+                return OperationCode::AtomicImageAdd;
+            case Tegra::Shader::ImageAtomicOperation::Min:
+                return OperationCode::AtomicImageMin;
+            case Tegra::Shader::ImageAtomicOperation::Max:
+                return OperationCode::AtomicImageMax;
+            case Tegra::Shader::ImageAtomicOperation::And:
+                return OperationCode::AtomicImageAnd;
+            case Tegra::Shader::ImageAtomicOperation::Or:
+                return OperationCode::AtomicImageOr;
+            case Tegra::Shader::ImageAtomicOperation::Xor:
+                return OperationCode::AtomicImageXor;
+            case Tegra::Shader::ImageAtomicOperation::Exch:
+                return OperationCode::AtomicImageExchange;
+            default:
+                UNIMPLEMENTED_MSG("Unimplemented operation={}",
+                                  static_cast<u32>(instr.suatom_d.operation.Value()));
+                return OperationCode::AtomicImageAdd;
+            }
+        }();
+
+        const auto& image{GetImage(instr.image, instr.suatom_d.image_type, instr.suatom_d.size)};
+        MetaImage meta{image, {std::move(value)}};
+        SetRegister(bb, instr.gpr0, Operation(operation_code, meta, std::move(coords)));
         break;
     }
     default:
@@ -77,38 +114,51 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
     return pc;
 }
 
-Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type) {
-    const auto offset{static_cast<u64>(image.index.Value())};
-
-    // If this image has already been used, return the existing mapping.
-    const auto it = used_images.find(offset);
-    if (it != used_images.end()) {
-        ASSERT(it->second.GetType() == type);
-        return it->second;
+Image& ShaderIR::GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type,
+                          std::optional<Tegra::Shader::ImageAtomicSize> size) {
+    const auto offset{static_cast<std::size_t>(image.index.Value())};
+    if (const auto image = TryUseExistingImage(offset, type, size)) {
+        return *image;
     }
 
-    // Otherwise create a new mapping for this image.
     const std::size_t next_index{used_images.size()};
-    return used_images.emplace(offset, Image{offset, next_index, type}).first->second;
+    return used_images.emplace(offset, Image{offset, next_index, type, size}).first->second;
 }
 
-Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type) {
+Image& ShaderIR::GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type,
+                                  std::optional<Tegra::Shader::ImageAtomicSize> size) {
     const Node image_register{GetRegister(reg)};
     const auto [base_image, cbuf_index, cbuf_offset]{
         TrackCbuf(image_register, global_code, static_cast<s64>(global_code.size()))};
     const auto cbuf_key{(static_cast<u64>(cbuf_index) << 32) | static_cast<u64>(cbuf_offset)};
 
-    // If this image has already been used, return the existing mapping.
-    const auto it = used_images.find(cbuf_key);
-    if (it != used_images.end()) {
-        ASSERT(it->second.GetType() == type);
-        return it->second;
+    if (const auto image = TryUseExistingImage(cbuf_key, type, size)) {
+        return *image;
     }
 
-    // Otherwise create a new mapping for this image.
     const std::size_t next_index{used_images.size()};
-    return used_images.emplace(cbuf_key, Image{cbuf_index, cbuf_offset, next_index, type})
+    return used_images.emplace(cbuf_key, Image{cbuf_index, cbuf_offset, next_index, type, size})
         .first->second;
 }
 
+Image* ShaderIR::TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type,
+                                     std::optional<Tegra::Shader::ImageAtomicSize> size) {
+    auto it = used_images.find(offset);
+    if (it == used_images.end()) {
+        return nullptr;
+    }
+    auto& image = it->second;
+    ASSERT(image.GetType() == type);
+
+    if (size) {
+        // We know the size, if it's known it has to be the same as before, otherwise we can set it.
+        if (image.IsSizeKnown()) {
+            ASSERT(image.GetSize() == size);
+        } else {
+            image.SetSize(*size);
+        }
+    }
+    return &image;
+}
+
 } // namespace VideoCommon::Shader

src/video_core/shader/decode/memory.cpp

@@ -35,7 +35,7 @@ u32 GetUniformTypeElementsCount(Tegra::Shader::UniformType uniform_type) {
         return 1;
     }
 }
-} // namespace
+} // Anonymous namespace
 
 u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
@@ -106,16 +106,17 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }
         break;
     }
-    case OpCode::Id::LD_L: {
-        LOG_DEBUG(HW_GPU, "LD_L cache management mode: {}",
-                  static_cast<u64>(instr.ld_l.unknown.Value()));
-
-        const auto GetLmem = [&](s32 offset) {
+    case OpCode::Id::LD_L:
+        LOG_DEBUG(HW_GPU, "LD_L cache management mode: {}", static_cast<u64>(instr.ld_l.unknown));
+        [[fallthrough]];
+    case OpCode::Id::LD_S: {
+        const auto GetMemory = [&](s32 offset) {
             ASSERT(offset % 4 == 0);
             const Node immediate_offset = Immediate(static_cast<s32>(instr.smem_imm) + offset);
             const Node address = Operation(OperationCode::IAdd, NO_PRECISE, GetRegister(instr.gpr8),
                                            immediate_offset);
-            return GetLocalMemory(address);
+            return opcode->get().GetId() == OpCode::Id::LD_S ? GetSharedMemory(address)
+                                                             : GetLocalMemory(address);
         };
 
         switch (instr.ldst_sl.type.Value()) {
@@ -135,14 +136,16 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
                     return 0;
                 }
             }();
-            for (u32 i = 0; i < count; ++i)
-                SetTemporary(bb, i, GetLmem(i * 4));
-            for (u32 i = 0; i < count; ++i)
+            for (u32 i = 0; i < count; ++i) {
+                SetTemporary(bb, i, GetMemory(i * 4));
+            }
+            for (u32 i = 0; i < count; ++i) {
                 SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
+            }
             break;
         }
         default:
-            UNIMPLEMENTED_MSG("LD_L Unhandled type: {}",
+            UNIMPLEMENTED_MSG("{} Unhandled type: {}", opcode->get().GetName(),
                               static_cast<u32>(instr.ldst_sl.type.Value()));
         }
         break;
@@ -209,27 +212,34 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
 
         break;
     }
-    case OpCode::Id::ST_L: {
+    case OpCode::Id::ST_L:
         LOG_DEBUG(HW_GPU, "ST_L cache management mode: {}",
                   static_cast<u64>(instr.st_l.cache_management.Value()));
-
-        const auto GetLmemAddr = [&](s32 offset) {
+        [[fallthrough]];
+    case OpCode::Id::ST_S: {
+        const auto GetAddress = [&](s32 offset) {
             ASSERT(offset % 4 == 0);
             const Node immediate = Immediate(static_cast<s32>(instr.smem_imm) + offset);
             return Operation(OperationCode::IAdd, NO_PRECISE, GetRegister(instr.gpr8), immediate);
         };
 
+        const auto set_memory = opcode->get().GetId() == OpCode::Id::ST_L
+                                    ? &ShaderIR::SetLocalMemory
+                                    : &ShaderIR::SetSharedMemory;
+
         switch (instr.ldst_sl.type.Value()) {
         case Tegra::Shader::StoreType::Bits128:
-            SetLocalMemory(bb, GetLmemAddr(12), GetRegister(instr.gpr0.Value() + 3));
-            SetLocalMemory(bb, GetLmemAddr(8), GetRegister(instr.gpr0.Value() + 2));
+            (this->*set_memory)(bb, GetAddress(12), GetRegister(instr.gpr0.Value() + 3));
+            (this->*set_memory)(bb, GetAddress(8), GetRegister(instr.gpr0.Value() + 2));
+            [[fallthrough]];
         case Tegra::Shader::StoreType::Bits64:
-            SetLocalMemory(bb, GetLmemAddr(4), GetRegister(instr.gpr0.Value() + 1));
+            (this->*set_memory)(bb, GetAddress(4), GetRegister(instr.gpr0.Value() + 1));
+            [[fallthrough]];
         case Tegra::Shader::StoreType::Bits32:
-            SetLocalMemory(bb, GetLmemAddr(0), GetRegister(instr.gpr0));
+            (this->*set_memory)(bb, GetAddress(0), GetRegister(instr.gpr0));
             break;
         default:
-            UNIMPLEMENTED_MSG("ST_L Unhandled type: {}",
+            UNIMPLEMENTED_MSG("{} unhandled type: {}", opcode->get().GetName(),
                               static_cast<u32>(instr.ldst_sl.type.Value()));
         }
         break;

src/video_core/shader/decode/shift.cpp

@@ -17,8 +17,8 @@ u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
 
-    const Node op_a = GetRegister(instr.gpr8);
-    const Node op_b = [&]() {
+    Node op_a = GetRegister(instr.gpr8);
+    Node op_b = [&]() {
         if (instr.is_b_imm) {
             return Immediate(instr.alu.GetSignedImm20_20());
         } else if (instr.is_b_gpr) {
@@ -32,16 +32,23 @@ u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
     case OpCode::Id::SHR_C:
     case OpCode::Id::SHR_R:
     case OpCode::Id::SHR_IMM: {
-        const Node value = SignedOperation(OperationCode::IArithmeticShiftRight,
-                                           instr.shift.is_signed, PRECISE, op_a, op_b);
+        if (instr.shr.wrap) {
+            op_b = Operation(OperationCode::UBitwiseAnd, std::move(op_b), Immediate(0x1f));
+        } else {
+            op_b = Operation(OperationCode::IMax, std::move(op_b), Immediate(0));
+            op_b = Operation(OperationCode::IMin, std::move(op_b), Immediate(31));
+        }
+
+        Node value = SignedOperation(OperationCode::IArithmeticShiftRight, instr.shift.is_signed,
+                                     std::move(op_a), std::move(op_b));
         SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
-        SetRegister(bb, instr.gpr0, value);
+        SetRegister(bb, instr.gpr0, std::move(value));
         break;
     }
     case OpCode::Id::SHL_C:
     case OpCode::Id::SHL_R:
     case OpCode::Id::SHL_IMM: {
-        const Node value = Operation(OperationCode::ILogicalShiftLeft, PRECISE, op_a, op_b);
+        const Node value = Operation(OperationCode::ILogicalShiftLeft, op_a, op_b);
         SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
         SetRegister(bb, instr.gpr0, value);
         break;

src/video_core/shader/decode/warp.cpp

@@ -13,6 +13,7 @@ namespace VideoCommon::Shader {
 using Tegra::Shader::Instruction;
 using Tegra::Shader::OpCode;
 using Tegra::Shader::Pred;
+using Tegra::Shader::ShuffleOperation;
 using Tegra::Shader::VoteOperation;
 
 namespace {
@@ -44,6 +45,52 @@ u32 ShaderIR::DecodeWarp(NodeBlock& bb, u32 pc) {
         SetPredicate(bb, instr.vote.dest_pred, vote);
         break;
     }
+    case OpCode::Id::SHFL: {
+        Node mask = instr.shfl.is_mask_imm ? Immediate(static_cast<u32>(instr.shfl.mask_imm))
+                                           : GetRegister(instr.gpr39);
+        Node width = [&] {
+            // Convert the obscure SHFL mask back into GL_NV_shader_thread_shuffle's width. This has
+            // been done reversing Nvidia's math. It won't work on all cases due to SHFL having
+            // different parameters that don't properly map to GLSL's interface, but it should work
+            // for cases emitted by Nvidia's compiler.
+            if (instr.shfl.operation == ShuffleOperation::Up) {
+                return Operation(
+                    OperationCode::ILogicalShiftRight,
+                    Operation(OperationCode::IAdd, std::move(mask), Immediate(-0x2000)),
+                    Immediate(8));
+            } else {
+                return Operation(OperationCode::ILogicalShiftRight,
+                                 Operation(OperationCode::IAdd, Immediate(0x201F),
+                                           Operation(OperationCode::INegate, std::move(mask))),
+                                 Immediate(8));
+            }
+        }();
+
+        const auto [operation, in_range] = [instr]() -> std::pair<OperationCode, OperationCode> {
+            switch (instr.shfl.operation) {
+            case ShuffleOperation::Idx:
+                return {OperationCode::ShuffleIndexed, OperationCode::InRangeShuffleIndexed};
+            case ShuffleOperation::Up:
+                return {OperationCode::ShuffleUp, OperationCode::InRangeShuffleUp};
+            case ShuffleOperation::Down:
+                return {OperationCode::ShuffleDown, OperationCode::InRangeShuffleDown};
+            case ShuffleOperation::Bfly:
+                return {OperationCode::ShuffleButterfly, OperationCode::InRangeShuffleButterfly};
+            }
+            UNREACHABLE_MSG("Invalid SHFL operation: {}",
+                            static_cast<u64>(instr.shfl.operation.Value()));
+            return {};
+        }();
+
+        // Setting the predicate before the register is intentional to avoid overwriting.
+        Node index = instr.shfl.is_index_imm ? Immediate(static_cast<u32>(instr.shfl.index_imm))
+                                             : GetRegister(instr.gpr20);
+        SetPredicate(bb, instr.shfl.pred48, Operation(in_range, index, width));
+        SetRegister(
+            bb, instr.gpr0,
+            Operation(operation, GetRegister(instr.gpr8), std::move(index), std::move(width)));
+        break;
+    }
     default:
         UNIMPLEMENTED_MSG("Unhandled warp instruction: {}", opcode->get().GetName());
         break;

src/video_core/shader/node.h

@@ -7,6 +7,7 @@
 #include <array>
 #include <cstddef>
 #include <memory>
+#include <optional>
 #include <string>
 #include <tuple>
 #include <utility>
@@ -148,7 +149,14 @@ enum class OperationCode {
     TextureQueryLod,        /// (MetaTexture, float[N] coords) -> float4
     TexelFetch,             /// (MetaTexture, int[N], int) -> float4
 
-    ImageStore, /// (MetaImage, float[N] coords) -> void
+    ImageStore,          /// (MetaImage, int[N] values) -> void
+    AtomicImageAdd,      /// (MetaImage, int[N] coords) -> void
+    AtomicImageMin,      /// (MetaImage, int[N] coords) -> void
+    AtomicImageMax,      /// (MetaImage, int[N] coords) -> void
+    AtomicImageAnd,      /// (MetaImage, int[N] coords) -> void
+    AtomicImageOr,       /// (MetaImage, int[N] coords) -> void
+    AtomicImageXor,      /// (MetaImage, int[N] coords) -> void
+    AtomicImageExchange, /// (MetaImage, int[N] coords) -> void
 
     Branch,         /// (uint branch_target) -> void
     BranchIndirect, /// (uint branch_target) -> void
@@ -173,6 +181,16 @@ enum class OperationCode {
     VoteAny,      /// (bool) -> bool
     VoteEqual,    /// (bool) -> bool
 
+    ShuffleIndexed,   /// (uint value, uint index, uint width) -> uint
+    ShuffleUp,        /// (uint value, uint index, uint width) -> uint
+    ShuffleDown,      /// (uint value, uint index, uint width) -> uint
+    ShuffleButterfly, /// (uint value, uint index, uint width) -> uint
+
+    InRangeShuffleIndexed,   /// (uint index, uint width) -> bool
+    InRangeShuffleUp,        /// (uint index, uint width) -> bool
+    InRangeShuffleDown,      /// (uint index, uint width) -> bool
+    InRangeShuffleButterfly, /// (uint index, uint width) -> bool
+
     Amount,
 };
 
@@ -198,12 +216,13 @@ class PredicateNode;
 class AbufNode;
 class CbufNode;
 class LmemNode;
+class SmemNode;
 class GmemNode;
 class CommentNode;
 
 using NodeData =
     std::variant<OperationNode, ConditionalNode, GprNode, ImmediateNode, InternalFlagNode,
-                 PredicateNode, AbufNode, CbufNode, LmemNode, GmemNode, CommentNode>;
+                 PredicateNode, AbufNode, CbufNode, LmemNode, SmemNode, GmemNode, CommentNode>;
 using Node = std::shared_ptr<NodeData>;
 using Node4 = std::array<Node, 4>;
 using NodeBlock = std::vector<Node>;
@@ -275,25 +294,32 @@ private:
 
 class Image final {
 public:
-    constexpr explicit Image(u64 offset, std::size_t index, Tegra::Shader::ImageType type)
-        : offset{offset}, index{index}, type{type}, is_bindless{false} {}
+    constexpr explicit Image(std::size_t offset, std::size_t index, Tegra::Shader::ImageType type,
+                             std::optional<Tegra::Shader::ImageAtomicSize> size)
+        : offset{offset}, index{index}, type{type}, is_bindless{false}, size{size} {}
 
     constexpr explicit Image(u32 cbuf_index, u32 cbuf_offset, std::size_t index,
-                             Tegra::Shader::ImageType type)
+                             Tegra::Shader::ImageType type,
+                             std::optional<Tegra::Shader::ImageAtomicSize> size)
         : offset{(static_cast<u64>(cbuf_index) << 32) | cbuf_offset}, index{index}, type{type},
-          is_bindless{true} {}
+          is_bindless{true}, size{size} {}
 
     constexpr explicit Image(std::size_t offset, std::size_t index, Tegra::Shader::ImageType type,
-                             bool is_bindless, bool is_written, bool is_read)
+                             bool is_bindless, bool is_written, bool is_read,
+                             std::optional<Tegra::Shader::ImageAtomicSize> size)
         : offset{offset}, index{index}, type{type}, is_bindless{is_bindless},
-          is_written{is_written}, is_read{is_read} {}
+          is_written{is_written}, is_read{is_read}, size{size} {}
+
+    void MarkWrite() {
+        is_written = true;
+    }
 
     void MarkRead() {
         is_read = true;
     }
 
-    void MarkWrite() {
-        is_written = true;
+    void SetSize(Tegra::Shader::ImageAtomicSize size_) {
+        size = size_;
     }
 
     constexpr std::size_t GetOffset() const {
@@ -312,25 +338,39 @@ public:
         return is_bindless;
     }
 
-    constexpr bool IsRead() const {
-        return is_read;
-    }
-
     constexpr bool IsWritten() const {
         return is_written;
     }
 
+    constexpr bool IsRead() const {
+        return is_read;
+    }
+
     constexpr std::pair<u32, u32> GetBindlessCBuf() const {
         return {static_cast<u32>(offset >> 32), static_cast<u32>(offset)};
     }
 
+    constexpr bool IsSizeKnown() const {
+        return size.has_value();
+    }
+
+    constexpr Tegra::Shader::ImageAtomicSize GetSize() const {
+        return size.value();
+    }
+
+    constexpr bool operator<(const Image& rhs) const {
+        return std::tie(offset, index, type, size, is_bindless) <
+               std::tie(rhs.offset, rhs.index, rhs.type, rhs.size, rhs.is_bindless);
+    }
+
 private:
     u64 offset{};
     std::size_t index{};
     Tegra::Shader::ImageType type{};
     bool is_bindless{};
-    bool is_read{};
     bool is_written{};
+    bool is_read{};
+    std::optional<Tegra::Shader::ImageAtomicSize> size{};
 };
 
 struct GlobalMemoryBase {
@@ -554,6 +594,19 @@ private:
     Node address;
 };
 
+/// Shared memory node
+class SmemNode final {
+public:
+    explicit SmemNode(Node address) : address{std::move(address)} {}
+
+    const Node& GetAddress() const {
+        return address;
+    }
+
+private:
+    Node address;
+};
+
 /// Global memory node
 class GmemNode final {
 public:

src/video_core/shader/shader_ir.cpp

@@ -137,6 +137,10 @@ Node ShaderIR::GetLocalMemory(Node address) {
     return MakeNode<LmemNode>(std::move(address));
 }
 
+Node ShaderIR::GetSharedMemory(Node address) {
+    return MakeNode<SmemNode>(std::move(address));
+}
+
 Node ShaderIR::GetTemporary(u32 id) {
     return GetRegister(Register::ZeroIndex + 1 + id);
 }
@@ -378,6 +382,11 @@ void ShaderIR::SetLocalMemory(NodeBlock& bb, Node address, Node value) {
         Operation(OperationCode::Assign, GetLocalMemory(std::move(address)), std::move(value)));
 }
 
+void ShaderIR::SetSharedMemory(NodeBlock& bb, Node address, Node value) {
+    bb.push_back(
+        Operation(OperationCode::Assign, GetSharedMemory(std::move(address)), std::move(value)));
+}
+
 void ShaderIR::SetTemporary(NodeBlock& bb, u32 id, Node value) {
     SetRegister(bb, Register::ZeroIndex + 1 + id, std::move(value));
 }

src/video_core/shader/shader_ir.h

@@ -208,6 +208,8 @@ private:
     Node GetInternalFlag(InternalFlag flag, bool negated = false);
     /// Generates a node representing a local memory address
     Node GetLocalMemory(Node address);
+    /// Generates a node representing a shared memory address
+    Node GetSharedMemory(Node address);
     /// Generates a temporary, internally it uses a post-RZ register
     Node GetTemporary(u32 id);
 
@@ -217,8 +219,10 @@ private:
     void SetPredicate(NodeBlock& bb, u64 dest, Node src);
     /// Sets an internal flag. src value must be a bool-evaluated node
     void SetInternalFlag(NodeBlock& bb, InternalFlag flag, Node value);
-    /// Sets a local memory address. address and value must be a number-evaluated node
+    /// Sets a local memory address with a value.
     void SetLocalMemory(NodeBlock& bb, Node address, Node value);
+    /// Sets a shared memory address with a value.
+    void SetSharedMemory(NodeBlock& bb, Node address, Node value);
     /// Sets a temporary. Internally it uses a post-RZ register
     void SetTemporary(NodeBlock& bb, u32 id, Node value);
 
@@ -272,10 +276,16 @@ private:
                                       bool is_shadow);
 
     /// Accesses an image.
-    Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type);
+    Image& GetImage(Tegra::Shader::Image image, Tegra::Shader::ImageType type,
+                    std::optional<Tegra::Shader::ImageAtomicSize> size = {});
 
     /// Access a bindless image sampler.
-    Image& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type);
+    Image& GetBindlessImage(Tegra::Shader::Register reg, Tegra::Shader::ImageType type,
+                            std::optional<Tegra::Shader::ImageAtomicSize> size = {});
+
+    /// Tries to access an existing image, updating it's state as needed
+    Image* TryUseExistingImage(u64 offset, Tegra::Shader::ImageType type,
+                               std::optional<Tegra::Shader::ImageAtomicSize> size);
 
     /// Extracts a sequence of bits from a node
     Node BitfieldExtract(Node value, u32 offset, u32 bits);

src/video_core/surface.cpp

@@ -513,6 +513,26 @@ bool IsPixelFormatASTC(PixelFormat format) {
     }
 }
 
+bool IsPixelFormatSRGB(PixelFormat format) {
+    switch (format) {
+    case PixelFormat::RGBA8_SRGB:
+    case PixelFormat::BGRA8_SRGB:
+    case PixelFormat::DXT1_SRGB:
+    case PixelFormat::DXT23_SRGB:
+    case PixelFormat::DXT45_SRGB:
+    case PixelFormat::BC7U_SRGB:
+    case PixelFormat::ASTC_2D_4X4_SRGB:
+    case PixelFormat::ASTC_2D_8X8_SRGB:
+    case PixelFormat::ASTC_2D_8X5_SRGB:
+    case PixelFormat::ASTC_2D_5X4_SRGB:
+    case PixelFormat::ASTC_2D_5X5_SRGB:
+    case PixelFormat::ASTC_2D_10X8_SRGB:
+        return true;
+    default:
+        return false;
+    }
+}
+
 std::pair<u32, u32> GetASTCBlockSize(PixelFormat format) {
     return {GetDefaultBlockWidth(format), GetDefaultBlockHeight(format)};
 }

src/video_core/surface.h

@@ -547,6 +547,8 @@ SurfaceType GetFormatType(PixelFormat pixel_format);
 
 bool IsPixelFormatASTC(PixelFormat format);
 
+bool IsPixelFormatSRGB(PixelFormat format);
+
 std::pair<u32, u32> GetASTCBlockSize(PixelFormat format);
 
 /// Returns true if the specified PixelFormat is a BCn format, e.g. DXT or DXN

src/yuzu/configuration/configure_input.cpp

@@ -182,6 +182,8 @@ void ConfigureInput::UpdateUIEnabled() {
         players_configure[i]->setEnabled(players_controller[i]->currentIndex() != 0);
     }
 
+    ui->handheld_connected->setChecked(ui->handheld_connected->isChecked() &&
+                                       !ui->use_docked_mode->isChecked());
     ui->handheld_connected->setEnabled(!ui->use_docked_mode->isChecked());
     ui->handheld_configure->setEnabled(ui->handheld_connected->isChecked() &&
                                        !ui->use_docked_mode->isChecked());