decode/image: Fix typo in assert in GetComponentSize()

The components' sizes were mismatched. This corrects that.

src/CMakeLists.txt

@@ -53,6 +53,7 @@ if (MSVC)
 else()
     add_compile_options(
         -Wall
+        -Werror=reorder
         -Wno-attributes
     )
 

src/core/file_sys/patch_manager.cpp

@@ -348,6 +348,12 @@ static void ApplyLayeredFS(VirtualFile& romfs, u64 title_id, ContentRecordType t
         if (ext_dir != nullptr)
             layers_ext.push_back(std::move(ext_dir));
     }
+
+    // When there are no layers to apply, return early as there is no need to rebuild the RomFS
+    if (layers.empty() && layers_ext.empty()) {
+        return;
+    }
+
     layers.push_back(std::move(extracted));
 
     auto layered = LayeredVfsDirectory::MakeLayeredDirectory(std::move(layers));

src/core/hle/kernel/kernel.cpp

@@ -103,7 +103,7 @@ static void ThreadWakeupCallback(u64 thread_handle, [[maybe_unused]] s64 cycles_
 
 struct KernelCore::Impl {
     explicit Impl(Core::System& system, KernelCore& kernel)
-        : system{system}, global_scheduler{kernel}, synchronization{system}, time_manager{system} {}
+        : global_scheduler{kernel}, synchronization{system}, time_manager{system}, system{system} {}
 
     void Initialize(KernelCore& kernel) {
         Shutdown();

src/core/hle/service/friend/friend.cpp

@@ -27,7 +27,7 @@ public:
             {10110, nullptr, "GetFriendProfileImage"},
             {10200, nullptr, "SendFriendRequestForApplication"},
             {10211, nullptr, "AddFacedFriendRequestForApplication"},
-            {10400, nullptr, "GetBlockedUserListIds"},
+            {10400, &IFriendService::GetBlockedUserListIds, "GetBlockedUserListIds"},
             {10500, nullptr, "GetProfileList"},
             {10600, nullptr, "DeclareOpenOnlinePlaySession"},
             {10601, &IFriendService::DeclareCloseOnlinePlaySession, "DeclareCloseOnlinePlaySession"},
@@ -121,6 +121,15 @@ private:
     };
     static_assert(sizeof(SizedFriendFilter) == 0x10, "SizedFriendFilter is an invalid size");
 
+    void GetBlockedUserListIds(Kernel::HLERequestContext& ctx) {
+        // This is safe to stub, as there should be no adverse consequences from reporting no
+        // blocked users.
+        LOG_WARNING(Service_ACC, "(STUBBED) called");
+        IPC::ResponseBuilder rb{ctx, 3};
+        rb.Push(RESULT_SUCCESS);
+        rb.Push<u32>(0); // Indicates there are no blocked users
+    }
+
     void DeclareCloseOnlinePlaySession(Kernel::HLERequestContext& ctx) {
         // Stub used by Splatoon 2
         LOG_WARNING(Service_ACC, "(STUBBED) called");

src/video_core/CMakeLists.txt

@@ -160,8 +160,6 @@ if (ENABLE_VULKAN)
         renderer_vulkan/fixed_pipeline_state.h
         renderer_vulkan/maxwell_to_vk.cpp
         renderer_vulkan/maxwell_to_vk.h
-        renderer_vulkan/nsight_aftermath_tracker.cpp
-        renderer_vulkan/nsight_aftermath_tracker.h
         renderer_vulkan/renderer_vulkan.h
         renderer_vulkan/renderer_vulkan.cpp
         renderer_vulkan/vk_blit_screen.cpp
@@ -215,30 +213,19 @@ if (ENABLE_VULKAN)
         renderer_vulkan/wrapper.cpp
         renderer_vulkan/wrapper.h
     )
+
+    target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include)
+    target_compile_definitions(video_core PRIVATE HAS_VULKAN)
 endif()
 
 create_target_directory_groups(video_core)
 
 target_link_libraries(video_core PUBLIC common core)
 target_link_libraries(video_core PRIVATE glad)
-
 if (ENABLE_VULKAN)
-    target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include)
-    target_compile_definitions(video_core PRIVATE HAS_VULKAN)
     target_link_libraries(video_core PRIVATE sirit)
 endif()
 
-if (ENABLE_NSIGHT_AFTERMATH)
-    if (NOT DEFINED ENV{NSIGHT_AFTERMATH_SDK})
-        message(ERROR "Environment variable NSIGHT_AFTERMATH_SDK has to be provided")
-    endif()
-    if (NOT WIN32)
-        message(ERROR "Nsight Aftermath doesn't support non-Windows platforms")
-    endif()
-    target_compile_definitions(video_core PRIVATE HAS_NSIGHT_AFTERMATH)
-    target_include_directories(video_core PRIVATE "$ENV{NSIGHT_AFTERMATH_SDK}/include")
-endif()
-
 if (MSVC)
     target_compile_options(video_core PRIVATE /we4267)
 else()

src/video_core/engines/maxwell_3d.h

@@ -303,6 +303,10 @@ public:
                 return (type == Type::SignedNorm) || (type == Type::UnsignedNorm);
             }
 
+            bool IsConstant() const {
+                return constant;
+            }
+
             bool IsValid() const {
                 return size != Size::Invalid;
             }

src/video_core/engines/shader_bytecode.h

@@ -1005,6 +1005,12 @@ union Instruction {
         BitField<46, 2, u64> cache_mode;
     } stg;
 
+    union {
+        BitField<23, 3, AtomicOp> operation;
+        BitField<48, 1, u64> extended;
+        BitField<20, 3, GlobalAtomicType> type;
+    } red;
+
     union {
         BitField<52, 4, AtomicOp> operation;
         BitField<49, 3, GlobalAtomicType> type;
@@ -1787,6 +1793,7 @@ public:
         ST_S,
         ST,    // Store in generic memory
         STG,   // Store in global memory
+        RED,   // Reduction operation
         ATOM,  // Atomic operation on global memory
         ATOMS, // Atomic operation on shared memory
         AL2P,  // Transforms attribute memory into physical memory
@@ -1871,7 +1878,8 @@ public:
         ICMP_R,
         ICMP_CR,
         ICMP_IMM,
-        FCMP_R,
+        FCMP_RR,
+        FCMP_RC,
         MUFU,  // Multi-Function Operator
         RRO_C, // Range Reduction Operator
         RRO_R,
@@ -2096,6 +2104,7 @@ private:
             INST("1110111101010---", Id::ST_L, Type::Memory, "ST_L"),
             INST("101-------------", Id::ST, Type::Memory, "ST"),
             INST("1110111011011---", Id::STG, Type::Memory, "STG"),
+            INST("1110101111111---", Id::RED, Type::Memory, "RED"),
             INST("11101101--------", Id::ATOM, Type::Memory, "ATOM"),
             INST("11101100--------", Id::ATOMS, Type::Memory, "ATOMS"),
             INST("1110111110100---", Id::AL2P, Type::Memory, "AL2P"),
@@ -2179,7 +2188,8 @@ private:
             INST("0101110100100---", Id::HSETP2_R, Type::HalfSetPredicate, "HSETP2_R"),
             INST("0111111-0-------", Id::HSETP2_IMM, Type::HalfSetPredicate, "HSETP2_IMM"),
             INST("0101110100011---", Id::HSET2_R, Type::HalfSet, "HSET2_R"),
-            INST("010110111010----", Id::FCMP_R, Type::Arithmetic, "FCMP_R"),
+            INST("010110111010----", Id::FCMP_RR, Type::Arithmetic, "FCMP_RR"),
+            INST("010010111010----", Id::FCMP_RC, Type::Arithmetic, "FCMP_RC"),
             INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"),
             INST("0100110010010---", Id::RRO_C, Type::Arithmetic, "RRO_C"),
             INST("0101110010010---", Id::RRO_R, Type::Arithmetic, "RRO_R"),

src/video_core/gpu_asynch.cpp

@@ -12,8 +12,9 @@ namespace VideoCommon {
 
 GPUAsynch::GPUAsynch(Core::System& system, std::unique_ptr<VideoCore::RendererBase>&& renderer_,
                      std::unique_ptr<Core::Frontend::GraphicsContext>&& context)
-    : GPU(system, std::move(renderer_), true), gpu_thread{system}, gpu_context(std::move(context)),
-      cpu_context(renderer->GetRenderWindow().CreateSharedContext()) {}
+    : GPU(system, std::move(renderer_), true), gpu_thread{system},
+      cpu_context(renderer->GetRenderWindow().CreateSharedContext()),
+      gpu_context(std::move(context)) {}
 
 GPUAsynch::~GPUAsynch() = default;
 

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -140,8 +140,8 @@ void RasterizerOpenGL::SetupVertexFormat() {
         const auto attrib = gpu.regs.vertex_attrib_format[index];
         const auto gl_index = static_cast<GLuint>(index);
 
-        // Ignore invalid attributes.
-        if (!attrib.IsValid()) {
+        // Disable constant attributes.
+        if (attrib.IsConstant()) {
             glDisableVertexAttribArray(gl_index);
             continue;
         }

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -34,6 +34,8 @@
 namespace OpenGL {
 
 using Tegra::Engines::ShaderType;
+using VideoCommon::Shader::CompileDepth;
+using VideoCommon::Shader::CompilerSettings;
 using VideoCommon::Shader::ProgramCode;
 using VideoCommon::Shader::Registry;
 using VideoCommon::Shader::ShaderIR;
@@ -43,7 +45,7 @@ namespace {
 constexpr u32 STAGE_MAIN_OFFSET = 10;
 constexpr u32 KERNEL_MAIN_OFFSET = 0;
 
-constexpr VideoCommon::Shader::CompilerSettings COMPILER_SETTINGS{};
+constexpr CompilerSettings COMPILER_SETTINGS{CompileDepth::FullDecompile};
 
 /// Gets the address for the specified shader stage program
 GPUVAddr GetShaderAddress(Core::System& system, Maxwell::ShaderProgram program) {

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -1821,13 +1821,15 @@ private:
     Expression HMergeH0(Operation operation) {
         const std::string dest = VisitOperand(operation, 0).AsUint();
         const std::string src = VisitOperand(operation, 1).AsUint();
-        return {fmt::format("bitfieldInsert({}, {}, 0, 16)", dest, src), Type::Uint};
+        return {fmt::format("vec2(unpackHalf2x16({}).x, unpackHalf2x16({}).y)", src, dest),
+                Type::HalfFloat};
     }
 
     Expression HMergeH1(Operation operation) {
         const std::string dest = VisitOperand(operation, 0).AsUint();
         const std::string src = VisitOperand(operation, 1).AsUint();
-        return {fmt::format("bitfieldInsert({}, {}, 16, 16)", dest, src), Type::Uint};
+        return {fmt::format("vec2(unpackHalf2x16({}).x, unpackHalf2x16({}).y)", dest, src),
+                Type::HalfFloat};
     }
 
     Expression HPack2(Operation operation) {
@@ -2117,8 +2119,14 @@ private:
             return {};
         }
         return {fmt::format("atomic{}({}, {})", opname, Visit(operation[0]).GetCode(),
-                            Visit(operation[1]).As(type)),
-                type};
+                            Visit(operation[1]).AsUint()),
+                Type::Uint};
+    }
+
+    template <const std::string_view& opname, Type type>
+    Expression Reduce(Operation operation) {
+        code.AddLine("{};", Atomic<opname, type>(operation).GetCode());
+        return {};
     }
 
     Expression Branch(Operation operation) {
@@ -2477,6 +2485,20 @@ private:
         &GLSLDecompiler::Atomic<Func::Or, Type::Int>,
         &GLSLDecompiler::Atomic<Func::Xor, Type::Int>,
 
+        &GLSLDecompiler::Reduce<Func::Add, Type::Uint>,
+        &GLSLDecompiler::Reduce<Func::Min, Type::Uint>,
+        &GLSLDecompiler::Reduce<Func::Max, Type::Uint>,
+        &GLSLDecompiler::Reduce<Func::And, Type::Uint>,
+        &GLSLDecompiler::Reduce<Func::Or, Type::Uint>,
+        &GLSLDecompiler::Reduce<Func::Xor, Type::Uint>,
+
+        &GLSLDecompiler::Reduce<Func::Add, Type::Int>,
+        &GLSLDecompiler::Reduce<Func::Min, Type::Int>,
+        &GLSLDecompiler::Reduce<Func::Max, Type::Int>,
+        &GLSLDecompiler::Reduce<Func::And, Type::Int>,
+        &GLSLDecompiler::Reduce<Func::Or, Type::Int>,
+        &GLSLDecompiler::Reduce<Func::Xor, Type::Int>,
+
         &GLSLDecompiler::Branch,
         &GLSLDecompiler::BranchIndirect,
         &GLSLDecompiler::PushFlowStack,

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -417,7 +417,7 @@ void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const {
 
         switch (params.target) {
         case SurfaceTarget::Texture2DArray:
-            glFramebufferTexture(target, attachment, GetTexture(), params.base_level);
+            glFramebufferTexture(target, attachment, GetTexture(), 0);
             break;
         default:
             UNIMPLEMENTED();

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -315,8 +315,8 @@ public:
 
 RendererOpenGL::RendererOpenGL(Core::Frontend::EmuWindow& emu_window, Core::System& system,
                                Core::Frontend::GraphicsContext& context)
-    : VideoCore::RendererBase{emu_window}, emu_window{emu_window}, system{system},
-      frame_mailbox{}, context{context}, has_debug_tool{HasDebugTool()} {}
+    : RendererBase{emu_window}, emu_window{emu_window}, system{system}, context{context},
+      has_debug_tool{HasDebugTool()} {}
 
 RendererOpenGL::~RendererOpenGL() = default;
 

src/video_core/renderer_vulkan/maxwell_to_vk.cpp

@@ -360,6 +360,7 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib
         default:
             break;
         }
+        break;
     case Maxwell::VertexAttribute::Type::UnsignedInt:
         switch (size) {
         case Maxwell::VertexAttribute::Size::Size_8:
@@ -370,6 +371,14 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib
             return VK_FORMAT_R8G8B8_UINT;
         case Maxwell::VertexAttribute::Size::Size_8_8_8_8:
             return VK_FORMAT_R8G8B8A8_UINT;
+        case Maxwell::VertexAttribute::Size::Size_16:
+            return VK_FORMAT_R16_UINT;
+        case Maxwell::VertexAttribute::Size::Size_16_16:
+            return VK_FORMAT_R16G16_UINT;
+        case Maxwell::VertexAttribute::Size::Size_16_16_16:
+            return VK_FORMAT_R16G16B16_UINT;
+        case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
+            return VK_FORMAT_R16G16B16A16_UINT;
         case Maxwell::VertexAttribute::Size::Size_32:
             return VK_FORMAT_R32_UINT;
         case Maxwell::VertexAttribute::Size::Size_32_32:
@@ -381,6 +390,7 @@ VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttrib
         default:
             break;
         }
+        break;
     case Maxwell::VertexAttribute::Type::UnsignedScaled:
         switch (size) {
         case Maxwell::VertexAttribute::Size::Size_8:

src/video_core/renderer_vulkan/nsight_aftermath_tracker.cpp

@@ -1,220 +0,0 @@
-// Copyright 2020 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#ifdef HAS_NSIGHT_AFTERMATH
-
-#include <mutex>
-#include <string>
-#include <string_view>
-#include <utility>
-#include <vector>
-
-#include <fmt/format.h>
-
-#define VK_NO_PROTOTYPES
-#include <vulkan/vulkan.h>
-
-#include <GFSDK_Aftermath.h>
-#include <GFSDK_Aftermath_Defines.h>
-#include <GFSDK_Aftermath_GpuCrashDump.h>
-#include <GFSDK_Aftermath_GpuCrashDumpDecoding.h>
-
-#include "common/common_paths.h"
-#include "common/common_types.h"
-#include "common/file_util.h"
-#include "common/logging/log.h"
-#include "common/scope_exit.h"
-
-#include "video_core/renderer_vulkan/nsight_aftermath_tracker.h"
-
-namespace Vulkan {
-
-static constexpr char AFTERMATH_LIB_NAME[] = "GFSDK_Aftermath_Lib.x64.dll";
-
-NsightAftermathTracker::NsightAftermathTracker() = default;
-
-NsightAftermathTracker::~NsightAftermathTracker() {
-    if (initialized) {
-        (void)GFSDK_Aftermath_DisableGpuCrashDumps();
-    }
-}
-
-bool NsightAftermathTracker::Initialize() {
-    if (!dl.Open(AFTERMATH_LIB_NAME)) {
-        LOG_ERROR(Render_Vulkan, "Failed to load Nsight Aftermath DLL");
-        return false;
-    }
-
-    if (!dl.GetSymbol("GFSDK_Aftermath_DisableGpuCrashDumps",
-                      &GFSDK_Aftermath_DisableGpuCrashDumps) ||
-        !dl.GetSymbol("GFSDK_Aftermath_EnableGpuCrashDumps",
-                      &GFSDK_Aftermath_EnableGpuCrashDumps) ||
-        !dl.GetSymbol("GFSDK_Aftermath_GetShaderDebugInfoIdentifier",
-                      &GFSDK_Aftermath_GetShaderDebugInfoIdentifier) ||
-        !dl.GetSymbol("GFSDK_Aftermath_GetShaderHashSpirv", &GFSDK_Aftermath_GetShaderHashSpirv) ||
-        !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_CreateDecoder",
-                      &GFSDK_Aftermath_GpuCrashDump_CreateDecoder) ||
-        !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_DestroyDecoder",
-                      &GFSDK_Aftermath_GpuCrashDump_DestroyDecoder) ||
-        !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_GenerateJSON",
-                      &GFSDK_Aftermath_GpuCrashDump_GenerateJSON) ||
-        !dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_GetJSON",
-                      &GFSDK_Aftermath_GpuCrashDump_GetJSON)) {
-        LOG_ERROR(Render_Vulkan, "Failed to load Nsight Aftermath function pointers");
-        return false;
-    }
-
-    dump_dir = FileUtil::GetUserPath(FileUtil::UserPath::LogDir) + "gpucrash";
-
-    (void)FileUtil::DeleteDirRecursively(dump_dir);
-    if (!FileUtil::CreateDir(dump_dir)) {
-        LOG_ERROR(Render_Vulkan, "Failed to create Nsight Aftermath dump directory");
-        return false;
-    }
-
-    if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_EnableGpuCrashDumps(
-            GFSDK_Aftermath_Version_API, GFSDK_Aftermath_GpuCrashDumpWatchedApiFlags_Vulkan,
-            GFSDK_Aftermath_GpuCrashDumpFeatureFlags_Default, GpuCrashDumpCallback,
-            ShaderDebugInfoCallback, CrashDumpDescriptionCallback, this))) {
-        LOG_ERROR(Render_Vulkan, "GFSDK_Aftermath_EnableGpuCrashDumps failed");
-        return false;
-    }
-
-    LOG_INFO(Render_Vulkan, "Nsight Aftermath dump directory is \"{}\"", dump_dir);
-
-    initialized = true;
-    return true;
-}
-
-void NsightAftermathTracker::SaveShader(const std::vector<u32>& spirv) const {
-    if (!initialized) {
-        return;
-    }
-
-    std::vector<u32> spirv_copy = spirv;
-    GFSDK_Aftermath_SpirvCode shader;
-    shader.pData = spirv_copy.data();
-    shader.size = static_cast<u32>(spirv_copy.size() * 4);
-
-    std::scoped_lock lock{mutex};
-
-    GFSDK_Aftermath_ShaderHash hash;
-    if (!GFSDK_Aftermath_SUCCEED(
-            GFSDK_Aftermath_GetShaderHashSpirv(GFSDK_Aftermath_Version_API, &shader, &hash))) {
-        LOG_ERROR(Render_Vulkan, "Failed to hash SPIR-V module");
-        return;
-    }
-
-    FileUtil::IOFile file(fmt::format("{}/source_{:016x}.spv", dump_dir, hash.hash), "wb");
-    if (!file.IsOpen()) {
-        LOG_ERROR(Render_Vulkan, "Failed to dump SPIR-V module with hash={:016x}", hash.hash);
-        return;
-    }
-    if (file.WriteArray(spirv.data(), spirv.size()) != spirv.size()) {
-        LOG_ERROR(Render_Vulkan, "Failed to write SPIR-V module with hash={:016x}", hash.hash);
-        return;
-    }
-}
-
-void NsightAftermathTracker::OnGpuCrashDumpCallback(const void* gpu_crash_dump,
-                                                    u32 gpu_crash_dump_size) {
-    std::scoped_lock lock{mutex};
-
-    LOG_CRITICAL(Render_Vulkan, "called");
-
-    GFSDK_Aftermath_GpuCrashDump_Decoder decoder;
-    if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GpuCrashDump_CreateDecoder(
-            GFSDK_Aftermath_Version_API, gpu_crash_dump, gpu_crash_dump_size, &decoder))) {
-        LOG_ERROR(Render_Vulkan, "Failed to create decoder");
-        return;
-    }
-    SCOPE_EXIT({ GFSDK_Aftermath_GpuCrashDump_DestroyDecoder(decoder); });
-
-    u32 json_size = 0;
-    if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GpuCrashDump_GenerateJSON(
-            decoder, GFSDK_Aftermath_GpuCrashDumpDecoderFlags_ALL_INFO,
-            GFSDK_Aftermath_GpuCrashDumpFormatterFlags_NONE, nullptr, nullptr, nullptr, nullptr,
-            this, &json_size))) {
-        LOG_ERROR(Render_Vulkan, "Failed to generate JSON");
-        return;
-    }
-    std::vector<char> json(json_size);
-    if (!GFSDK_Aftermath_SUCCEED(
-            GFSDK_Aftermath_GpuCrashDump_GetJSON(decoder, json_size, json.data()))) {
-        LOG_ERROR(Render_Vulkan, "Failed to query JSON");
-        return;
-    }
-
-    const std::string base_name = [this] {
-        const int id = dump_id++;
-        if (id == 0) {
-            return fmt::format("{}/crash.nv-gpudmp", dump_dir);
-        } else {
-            return fmt::format("{}/crash_{}.nv-gpudmp", dump_dir, id);
-        }
-    }();
-
-    std::string_view dump_view(static_cast<const char*>(gpu_crash_dump), gpu_crash_dump_size);
-    if (FileUtil::WriteStringToFile(false, base_name, dump_view) != gpu_crash_dump_size) {
-        LOG_ERROR(Render_Vulkan, "Failed to write dump file");
-        return;
-    }
-    const std::string_view json_view(json.data(), json.size());
-    if (FileUtil::WriteStringToFile(true, base_name + ".json", json_view) != json.size()) {
-        LOG_ERROR(Render_Vulkan, "Failed to write JSON");
-        return;
-    }
-}
-
-void NsightAftermathTracker::OnShaderDebugInfoCallback(const void* shader_debug_info,
-                                                       u32 shader_debug_info_size) {
-    std::scoped_lock lock{mutex};
-
-    GFSDK_Aftermath_ShaderDebugInfoIdentifier identifier;
-    if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GetShaderDebugInfoIdentifier(
-            GFSDK_Aftermath_Version_API, shader_debug_info, shader_debug_info_size, &identifier))) {
-        LOG_ERROR(Render_Vulkan, "GFSDK_Aftermath_GetShaderDebugInfoIdentifier failed");
-        return;
-    }
-
-    const std::string path =
-        fmt::format("{}/shader_{:016x}{:016x}.nvdbg", dump_dir, identifier.id[0], identifier.id[1]);
-    FileUtil::IOFile file(path, "wb");
-    if (!file.IsOpen()) {
-        LOG_ERROR(Render_Vulkan, "Failed to create file {}", path);
-        return;
-    }
-    if (file.WriteBytes(static_cast<const u8*>(shader_debug_info), shader_debug_info_size) !=
-        shader_debug_info_size) {
-        LOG_ERROR(Render_Vulkan, "Failed to write file {}", path);
-        return;
-    }
-}
-
-void NsightAftermathTracker::OnCrashDumpDescriptionCallback(
-    PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description) {
-    add_description(GFSDK_Aftermath_GpuCrashDumpDescriptionKey_ApplicationName, "yuzu");
-}
-
-void NsightAftermathTracker::GpuCrashDumpCallback(const void* gpu_crash_dump,
-                                                  u32 gpu_crash_dump_size, void* user_data) {
-    static_cast<NsightAftermathTracker*>(user_data)->OnGpuCrashDumpCallback(gpu_crash_dump,
-                                                                            gpu_crash_dump_size);
-}
-
-void NsightAftermathTracker::ShaderDebugInfoCallback(const void* shader_debug_info,
-                                                     u32 shader_debug_info_size, void* user_data) {
-    static_cast<NsightAftermathTracker*>(user_data)->OnShaderDebugInfoCallback(
-        shader_debug_info, shader_debug_info_size);
-}
-
-void NsightAftermathTracker::CrashDumpDescriptionCallback(
-    PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description, void* user_data) {
-    static_cast<NsightAftermathTracker*>(user_data)->OnCrashDumpDescriptionCallback(
-        add_description);
-}
-
-} // namespace Vulkan
-
-#endif // HAS_NSIGHT_AFTERMATH

src/video_core/renderer_vulkan/nsight_aftermath_tracker.h

@@ -1,87 +0,0 @@
-// Copyright 2020 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <mutex>
-#include <string>
-#include <vector>
-
-#define VK_NO_PROTOTYPES
-#include <vulkan/vulkan.h>
-
-#ifdef HAS_NSIGHT_AFTERMATH
-#include <GFSDK_Aftermath_Defines.h>
-#include <GFSDK_Aftermath_GpuCrashDump.h>
-#include <GFSDK_Aftermath_GpuCrashDumpDecoding.h>
-#endif
-
-#include "common/common_types.h"
-#include "common/dynamic_library.h"
-
-namespace Vulkan {
-
-class NsightAftermathTracker {
-public:
-    NsightAftermathTracker();
-    ~NsightAftermathTracker();
-
-    NsightAftermathTracker(const NsightAftermathTracker&) = delete;
-    NsightAftermathTracker& operator=(const NsightAftermathTracker&) = delete;
-
-    // Delete move semantics because Aftermath initialization uses a pointer to this.
-    NsightAftermathTracker(NsightAftermathTracker&&) = delete;
-    NsightAftermathTracker& operator=(NsightAftermathTracker&&) = delete;
-
-    bool Initialize();
-
-    void SaveShader(const std::vector<u32>& spirv) const;
-
-private:
-#ifdef HAS_NSIGHT_AFTERMATH
-    static void GpuCrashDumpCallback(const void* gpu_crash_dump, u32 gpu_crash_dump_size,
-                                     void* user_data);
-
-    static void ShaderDebugInfoCallback(const void* shader_debug_info, u32 shader_debug_info_size,
-                                        void* user_data);
-
-    static void CrashDumpDescriptionCallback(
-        PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description, void* user_data);
-
-    void OnGpuCrashDumpCallback(const void* gpu_crash_dump, u32 gpu_crash_dump_size);
-
-    void OnShaderDebugInfoCallback(const void* shader_debug_info, u32 shader_debug_info_size);
-
-    void OnCrashDumpDescriptionCallback(
-        PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description);
-
-    mutable std::mutex mutex;
-
-    std::string dump_dir;
-    int dump_id = 0;
-
-    bool initialized = false;
-
-    Common::DynamicLibrary dl;
-    PFN_GFSDK_Aftermath_DisableGpuCrashDumps GFSDK_Aftermath_DisableGpuCrashDumps;
-    PFN_GFSDK_Aftermath_EnableGpuCrashDumps GFSDK_Aftermath_EnableGpuCrashDumps;
-    PFN_GFSDK_Aftermath_GetShaderDebugInfoIdentifier GFSDK_Aftermath_GetShaderDebugInfoIdentifier;
-    PFN_GFSDK_Aftermath_GetShaderHashSpirv GFSDK_Aftermath_GetShaderHashSpirv;
-    PFN_GFSDK_Aftermath_GpuCrashDump_CreateDecoder GFSDK_Aftermath_GpuCrashDump_CreateDecoder;
-    PFN_GFSDK_Aftermath_GpuCrashDump_DestroyDecoder GFSDK_Aftermath_GpuCrashDump_DestroyDecoder;
-    PFN_GFSDK_Aftermath_GpuCrashDump_GenerateJSON GFSDK_Aftermath_GpuCrashDump_GenerateJSON;
-    PFN_GFSDK_Aftermath_GpuCrashDump_GetJSON GFSDK_Aftermath_GpuCrashDump_GetJSON;
-#endif
-};
-
-#ifndef HAS_NSIGHT_AFTERMATH
-inline NsightAftermathTracker::NsightAftermathTracker() = default;
-inline NsightAftermathTracker::~NsightAftermathTracker() = default;
-inline bool NsightAftermathTracker::Initialize() {
-    return false;
-}
-inline void NsightAftermathTracker::SaveShader(const std::vector<u32>&) const {}
-#endif
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_blit_screen.cpp

@@ -535,7 +535,9 @@ void VKBlitScreen::CreateGraphicsPipeline() {
     viewport_state_ci.pNext = nullptr;
     viewport_state_ci.flags = 0;
     viewport_state_ci.viewportCount = 1;
+    viewport_state_ci.pViewports = nullptr;
     viewport_state_ci.scissorCount = 1;
+    viewport_state_ci.pScissors = nullptr;
 
     VkPipelineRasterizationStateCreateInfo rasterization_ci;
     rasterization_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO;

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -105,8 +105,6 @@ vk::DescriptorUpdateTemplateKHR VKComputePipeline::CreateDescriptorUpdateTemplat
 }
 
 vk::ShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const {
-    device.SaveShader(code);
-
     VkShaderModuleCreateInfo ci;
     ci.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
     ci.pNext = nullptr;

src/video_core/renderer_vulkan/vk_device.cpp

@@ -9,7 +9,6 @@
 #include <string_view>
 #include <thread>
 #include <unordered_set>
-#include <utility>
 #include <vector>
 
 #include "common/assert.h"
@@ -168,7 +167,6 @@ bool VKDevice::Create() {
     VkPhysicalDeviceFeatures2 features2;
     features2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
     features2.pNext = nullptr;
-    const void* first_next = &features2;
     void** next = &features2.pNext;
 
     auto& features = features2.features;
@@ -298,19 +296,7 @@ bool VKDevice::Create() {
         LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted");
     }
 
-    VkDeviceDiagnosticsConfigCreateInfoNV diagnostics_nv;
-    if (nv_device_diagnostics_config) {
-        nsight_aftermath_tracker.Initialize();
-
-        diagnostics_nv.sType = VK_STRUCTURE_TYPE_DEVICE_DIAGNOSTICS_CONFIG_CREATE_INFO_NV;
-        diagnostics_nv.pNext = &features2;
-        diagnostics_nv.flags = VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_SHADER_DEBUG_INFO_BIT_NV |
-                               VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_RESOURCE_TRACKING_BIT_NV |
-                               VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_AUTOMATIC_CHECKPOINTS_BIT_NV;
-        first_next = &diagnostics_nv;
-    }
-
-    logical = vk::Device::Create(physical, queue_cis, extensions, first_next, dld);
+    logical = vk::Device::Create(physical, queue_cis, extensions, features2, dld);
     if (!logical) {
         LOG_ERROR(Render_Vulkan, "Failed to create logical device");
         return false;
@@ -358,12 +344,17 @@ VkFormat VKDevice::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFla
 void VKDevice::ReportLoss() const {
     LOG_CRITICAL(Render_Vulkan, "Device loss occured!");
 
-    // Wait for the log to flush and for Nsight Aftermath to dump the results
-    std::this_thread::sleep_for(std::chrono::seconds{3});
-}
+    // Wait some time to let the log flush
+    std::this_thread::sleep_for(std::chrono::seconds{1});
 
-void VKDevice::SaveShader(const std::vector<u32>& spirv) const {
-    nsight_aftermath_tracker.SaveShader(spirv);
+    if (!nv_device_diagnostic_checkpoints) {
+        return;
+    }
+
+    [[maybe_unused]] const std::vector data = graphics_queue.GetCheckpointDataNV(dld);
+    // Catch here in debug builds (or with optimizations disabled) the last graphics pipeline to be
+    // executed. It can be done on a debugger by evaluating the expression:
+    // *(VKGraphicsPipeline*)data[0]
 }
 
 bool VKDevice::IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features) const {
@@ -536,8 +527,8 @@ std::vector<const char*> VKDevice::LoadExtensions() {
         Test(extension, has_ext_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME,
              false);
         if (Settings::values.renderer_debug) {
-            Test(extension, nv_device_diagnostics_config,
-                 VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME, true);
+            Test(extension, nv_device_diagnostic_checkpoints,
+                 VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME, true);
         }
     }
 

src/video_core/renderer_vulkan/vk_device.h

@@ -10,7 +10,6 @@
 #include <vector>
 
 #include "common/common_types.h"
-#include "video_core/renderer_vulkan/nsight_aftermath_tracker.h"
 #include "video_core/renderer_vulkan/wrapper.h"
 
 namespace Vulkan {
@@ -44,9 +43,6 @@ public:
     /// Reports a device loss.
     void ReportLoss() const;
 
-    /// Reports a shader to Nsight Aftermath.
-    void SaveShader(const std::vector<u32>& spirv) const;
-
     /// Returns the dispatch loader with direct function pointers of the device.
     const vk::DeviceDispatch& GetDispatchLoader() const {
         return dld;
@@ -177,6 +173,11 @@ public:
         return ext_transform_feedback;
     }
 
+    /// Returns true if the device supports VK_NV_device_diagnostic_checkpoints.
+    bool IsNvDeviceDiagnosticCheckpoints() const {
+        return nv_device_diagnostic_checkpoints;
+    }
+
     /// Returns the vendor name reported from Vulkan.
     std::string_view GetVendorName() const {
         return vendor_name;
@@ -232,7 +233,7 @@ private:
     bool ext_depth_range_unrestricted{};       ///< Support for VK_EXT_depth_range_unrestricted.
     bool ext_shader_viewport_index_layer{};    ///< Support for VK_EXT_shader_viewport_index_layer.
     bool ext_transform_feedback{};             ///< Support for VK_EXT_transform_feedback.
-    bool nv_device_diagnostics_config{};       ///< Support for VK_NV_device_diagnostics_config.
+    bool nv_device_diagnostic_checkpoints{};   ///< Support for VK_NV_device_diagnostic_checkpoints.
 
     // Telemetry parameters
     std::string vendor_name;                      ///< Device's driver name.
@@ -240,9 +241,6 @@ private:
 
     /// Format properties dictionary.
     std::unordered_map<VkFormat, VkFormatProperties> format_properties;
-
-    /// Nsight Aftermath GPU crash tracker
-    NsightAftermathTracker nsight_aftermath_tracker;
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -147,8 +147,6 @@ std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules(
             continue;
         }
 
-        device.SaveShader(stage->code);
-
         ci.codeSize = stage->code.size() * sizeof(u32);
         ci.pCode = stage->code.data();
         modules.push_back(device.GetLogical().CreateShaderModule(ci));

src/video_core/renderer_vulkan/vk_query_cache.cpp

@@ -113,19 +113,8 @@ u64 HostCounter::BlockingQuery() const {
     if (ticks >= cache.Scheduler().Ticks()) {
         cache.Scheduler().Flush();
     }
-    u64 data;
-    const VkResult result = cache.Device().GetLogical().GetQueryResults(
-        query.first, query.second, 1, sizeof(data), &data, sizeof(data),
-        VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
-    switch (result) {
-    case VK_SUCCESS:
-        return data;
-    case VK_ERROR_DEVICE_LOST:
-        cache.Device().ReportLoss();
-        [[fallthrough]];
-    default:
-        throw vk::Exception(result);
-    }
+    return cache.Device().GetLogical().GetQueryResult<u64>(
+        query.first, query.second, VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -62,13 +62,16 @@ constexpr auto ComputeShaderIndex = static_cast<std::size_t>(Tegra::Engines::Sha
 
 VkViewport GetViewportState(const VKDevice& device, const Maxwell& regs, std::size_t index) {
     const auto& src = regs.viewport_transform[index];
+    const float width = src.scale_x * 2.0f;
+    const float height = src.scale_y * 2.0f;
+
     VkViewport viewport;
     viewport.x = src.translate_x - src.scale_x;
     viewport.y = src.translate_y - src.scale_y;
-    viewport.width = src.scale_x * 2.0f;
-    viewport.height = src.scale_y * 2.0f;
+    viewport.width = width != 0.0f ? width : 1.0f;
+    viewport.height = height != 0.0f ? height : 1.0f;
 
-    const float reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne;
+    const float reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1.0f : 0.0f;
     viewport.minDepth = src.translate_z - src.scale_z * reduce_z;
     viewport.maxDepth = src.translate_z + src.scale_z;
     if (!device.IsExtDepthRangeUnrestrictedSupported()) {
@@ -347,6 +350,11 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
 
     buffer_bindings.Bind(scheduler);
 
+    if (device.IsNvDeviceDiagnosticCheckpoints()) {
+        scheduler.Record(
+            [&pipeline](vk::CommandBuffer cmdbuf) { cmdbuf.SetCheckpointNV(&pipeline); });
+    }
+
     BeginTransformFeedback();
 
     const auto pipeline_layout = pipeline.GetLayout();
@@ -473,6 +481,11 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
     TransitionImages(image_views, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
                      VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT);
 
+    if (device.IsNvDeviceDiagnosticCheckpoints()) {
+        scheduler.Record(
+            [&pipeline](vk::CommandBuffer cmdbuf) { cmdbuf.SetCheckpointNV(nullptr); });
+    }
+
     scheduler.Record([grid_x = launch_desc.grid_dim_x, grid_y = launch_desc.grid_dim_y,
                       grid_z = launch_desc.grid_dim_z, pipeline_handle = pipeline.GetHandle(),
                       layout = pipeline.GetLayout(),

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -166,15 +166,7 @@ void VKScheduler::SubmitExecution(VkSemaphore semaphore) {
     submit_info.pCommandBuffers = current_cmdbuf.address();
     submit_info.signalSemaphoreCount = semaphore ? 1 : 0;
     submit_info.pSignalSemaphores = &semaphore;
-    switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info, *current_fence)) {
-    case VK_SUCCESS:
-        break;
-    case VK_ERROR_DEVICE_LOST:
-        device.ReportLoss();
-        [[fallthrough]];
-    default:
-        vk::Check(result);
-    }
+    device.GetGraphicsQueue().Submit(submit_info, *current_fence);
 }
 
 void VKScheduler::AllocateNewContext() {

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -1938,11 +1938,8 @@ private:
         return {};
     }
 
-    template <Id (Module::*func)(Id, Id, Id, Id, Id), Type result_type,
-              Type value_type = result_type>
+    template <Id (Module::*func)(Id, Id, Id, Id, Id)>
     Expression Atomic(Operation operation) {
-        const Id type_def = GetTypeDefinition(result_type);
-
         Id pointer;
         if (const auto smem = std::get_if<SmemNode>(&*operation[0])) {
             pointer = GetSharedMemoryPointer(*smem);
@@ -1950,15 +1947,19 @@ private:
             pointer = GetGlobalMemoryPointer(*gmem);
         } else {
             UNREACHABLE();
-            return {Constant(type_def, 0), result_type};
+            return {v_float_zero, Type::Float};
         }
-
-        const Id value = As(Visit(operation[1]), value_type);
-
         const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
-        const Id semantics = Constant(type_def, 0);
+        const Id semantics = Constant(t_uint, 0);
+        const Id value = AsUint(Visit(operation[1]));
 
-        return {(this->*func)(type_def, pointer, scope, semantics, value), result_type};
+        return {(this->*func)(t_uint, pointer, scope, semantics, value), Type::Uint};
+    }
+
+    template <Id (Module::*func)(Id, Id, Id, Id, Id)>
+    Expression Reduce(Operation operation) {
+        Atomic<func>(operation);
+        return {};
     }
 
     Expression Branch(Operation operation) {
@@ -2547,21 +2548,35 @@ private:
         &SPIRVDecompiler::AtomicImageXor,
         &SPIRVDecompiler::AtomicImageExchange,
 
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicExchange, Type::Uint>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd, Type::Uint>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicUMin, Type::Uint>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicUMax, Type::Uint>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicAnd, Type::Uint>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicOr, Type::Uint>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicXor, Type::Uint>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicExchange>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicUMin>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicUMax>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicAnd>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicOr>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicXor>,
 
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicExchange, Type::Int>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd, Type::Int>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicSMin, Type::Int>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicSMax, Type::Int>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicAnd, Type::Int>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicOr, Type::Int>,
-        &SPIRVDecompiler::Atomic<&Module::OpAtomicXor, Type::Int>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicExchange>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicIAdd>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicSMin>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicSMax>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicAnd>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicOr>,
+        &SPIRVDecompiler::Atomic<&Module::OpAtomicXor>,
+
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicIAdd>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicUMin>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicUMax>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicAnd>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicOr>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicXor>,
+
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicIAdd>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicSMin>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicSMax>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicAnd>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicOr>,
+        &SPIRVDecompiler::Reduce<&Module::OpAtomicXor>,
 
         &SPIRVDecompiler::Branch,
         &SPIRVDecompiler::BranchIndirect,

src/video_core/renderer_vulkan/wrapper.cpp

@@ -61,6 +61,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkCmdPipelineBarrier);
     X(vkCmdPushConstants);
     X(vkCmdSetBlendConstants);
+    X(vkCmdSetCheckpointNV);
     X(vkCmdSetDepthBias);
     X(vkCmdSetDepthBounds);
     X(vkCmdSetScissor);
@@ -115,6 +116,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkGetFenceStatus);
     X(vkGetImageMemoryRequirements);
     X(vkGetQueryPoolResults);
+    X(vkGetQueueCheckpointDataNV);
     X(vkMapMemory);
     X(vkQueueSubmit);
     X(vkResetFences);
@@ -407,6 +409,17 @@ DebugCallback Instance::TryCreateDebugCallback(
     return DebugCallback(messenger, handle, *dld);
 }
 
+std::vector<VkCheckpointDataNV> Queue::GetCheckpointDataNV(const DeviceDispatch& dld) const {
+    if (!dld.vkGetQueueCheckpointDataNV) {
+        return {};
+    }
+    u32 num;
+    dld.vkGetQueueCheckpointDataNV(queue, &num, nullptr);
+    std::vector<VkCheckpointDataNV> checkpoints(num);
+    dld.vkGetQueueCheckpointDataNV(queue, &num, checkpoints.data());
+    return checkpoints;
+}
+
 void Buffer::BindMemory(VkDeviceMemory memory, VkDeviceSize offset) const {
     Check(dld->vkBindBufferMemory(owner, handle, memory, offset));
 }
@@ -456,11 +469,12 @@ std::vector<VkImage> SwapchainKHR::GetImages() const {
 }
 
 Device Device::Create(VkPhysicalDevice physical_device, Span<VkDeviceQueueCreateInfo> queues_ci,
-                      Span<const char*> enabled_extensions, const void* next,
+                      Span<const char*> enabled_extensions,
+                      const VkPhysicalDeviceFeatures2& enabled_features,
                       DeviceDispatch& dld) noexcept {
     VkDeviceCreateInfo ci;
     ci.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO;
-    ci.pNext = next;
+    ci.pNext = &enabled_features;
     ci.flags = 0;
     ci.queueCreateInfoCount = queues_ci.size();
     ci.pQueueCreateInfos = queues_ci.data();

src/video_core/renderer_vulkan/wrapper.h

@@ -197,6 +197,7 @@ struct DeviceDispatch : public InstanceDispatch {
     PFN_vkCmdPipelineBarrier vkCmdPipelineBarrier;
     PFN_vkCmdPushConstants vkCmdPushConstants;
     PFN_vkCmdSetBlendConstants vkCmdSetBlendConstants;
+    PFN_vkCmdSetCheckpointNV vkCmdSetCheckpointNV;
     PFN_vkCmdSetDepthBias vkCmdSetDepthBias;
     PFN_vkCmdSetDepthBounds vkCmdSetDepthBounds;
     PFN_vkCmdSetScissor vkCmdSetScissor;
@@ -251,6 +252,7 @@ struct DeviceDispatch : public InstanceDispatch {
     PFN_vkGetFenceStatus vkGetFenceStatus;
     PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements;
     PFN_vkGetQueryPoolResults vkGetQueryPoolResults;
+    PFN_vkGetQueueCheckpointDataNV vkGetQueueCheckpointDataNV;
     PFN_vkMapMemory vkMapMemory;
     PFN_vkQueueSubmit vkQueueSubmit;
     PFN_vkResetFences vkResetFences;
@@ -565,8 +567,12 @@ public:
     /// Construct a queue handle.
     constexpr Queue(VkQueue queue, const DeviceDispatch& dld) noexcept : queue{queue}, dld{&dld} {}
 
-    VkResult Submit(Span<VkSubmitInfo> submit_infos, VkFence fence) const noexcept {
-        return dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence);
+    /// Returns the checkpoint data.
+    /// @note Returns an empty vector when the function pointer is not present.
+    std::vector<VkCheckpointDataNV> GetCheckpointDataNV(const DeviceDispatch& dld) const;
+
+    void Submit(Span<VkSubmitInfo> submit_infos, VkFence fence) const {
+        Check(dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence));
     }
 
     VkResult Present(const VkPresentInfoKHR& present_info) const noexcept {
@@ -653,7 +659,8 @@ class Device : public Handle<VkDevice, NoOwner, DeviceDispatch> {
 
 public:
     static Device Create(VkPhysicalDevice physical_device, Span<VkDeviceQueueCreateInfo> queues_ci,
-                         Span<const char*> enabled_extensions, const void* next,
+                         Span<const char*> enabled_extensions,
+                         const VkPhysicalDeviceFeatures2& enabled_features,
                          DeviceDispatch& dld) noexcept;
 
     Queue GetQueue(u32 family_index) const noexcept;
@@ -727,11 +734,18 @@ public:
         dld->vkResetQueryPoolEXT(handle, query_pool, first, count);
     }
 
-    VkResult GetQueryResults(VkQueryPool query_pool, u32 first, u32 count, std::size_t data_size,
-                             void* data, VkDeviceSize stride, VkQueryResultFlags flags) const
-        noexcept {
-        return dld->vkGetQueryPoolResults(handle, query_pool, first, count, data_size, data, stride,
-                                          flags);
+    void GetQueryResults(VkQueryPool query_pool, u32 first, u32 count, std::size_t data_size,
+                         void* data, VkDeviceSize stride, VkQueryResultFlags flags) const {
+        Check(dld->vkGetQueryPoolResults(handle, query_pool, first, count, data_size, data, stride,
+                                         flags));
+    }
+
+    template <typename T>
+    T GetQueryResult(VkQueryPool query_pool, u32 first, VkQueryResultFlags flags) const {
+        static_assert(std::is_trivially_copyable_v<T>);
+        T value;
+        GetQueryResults(query_pool, first, 1, sizeof(T), &value, sizeof(T), flags);
+        return value;
     }
 };
 
@@ -906,6 +920,10 @@ public:
         dld->vkCmdPushConstants(handle, layout, flags, offset, size, values);
     }
 
+    void SetCheckpointNV(const void* checkpoint_marker) const noexcept {
+        dld->vkCmdSetCheckpointNV(handle, checkpoint_marker);
+    }
+
     void SetViewport(u32 first, Span<VkViewport> viewports) const noexcept {
         dld->vkCmdSetViewport(handle, first, viewports.size(), viewports.data());
     }

src/video_core/shader/decode/arithmetic.cpp

@@ -136,7 +136,8 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) {
         SetRegister(bb, instr.gpr0, value);
         break;
     }
-    case OpCode::Id::FCMP_R: {
+    case OpCode::Id::FCMP_RR:
+    case OpCode::Id::FCMP_RC: {
         UNIMPLEMENTED_IF(instr.fcmp.ftz == 0);
         Node op_c = GetRegister(instr.gpr39);
         Node comp = GetPredicateComparisonFloat(instr.fcmp.cond, std::move(op_c), Immediate(0.0f));

src/video_core/shader/decode/image.cpp

@@ -119,7 +119,7 @@ ComponentType GetComponentType(Tegra::Engines::SamplerDescriptor descriptor,
         }
         break;
     }
-    UNIMPLEMENTED_MSG("texture format not implement={}", format);
+    UNIMPLEMENTED_MSG("Texture format not implemented={}", format);
     return ComponentType::FLOAT;
 }
 
@@ -201,10 +201,10 @@ u32 GetComponentSize(TextureFormat format, std::size_t component) {
         return 0;
     case TextureFormat::G24R8:
         if (component == 0) {
-            return 8;
+            return 24;
         }
         if (component == 1) {
-            return 24;
+            return 8;
         }
         return 0;
     case TextureFormat::G8R8:
@@ -212,7 +212,7 @@ u32 GetComponentSize(TextureFormat format, std::size_t component) {
     case TextureFormat::G4R4:
         return (component == 0 || component == 1) ? 4 : 0;
     default:
-        UNIMPLEMENTED_MSG("texture format not implement={}", format);
+        UNIMPLEMENTED_MSG("Texture format not implemented={}", format);
         return 0;
     }
 }
@@ -249,7 +249,7 @@ std::size_t GetImageComponentMask(TextureFormat format) {
     case TextureFormat::R1:
         return std::size_t{R};
     default:
-        UNIMPLEMENTED_MSG("texture format not implement={}", format);
+        UNIMPLEMENTED_MSG("Texture format not implemented={}", format);
         return std::size_t{R | G | B | A};
     }
 }

src/video_core/shader/decode/memory.cpp

@@ -3,7 +3,9 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
+#include <utility>
 #include <vector>
+
 #include <fmt/format.h>
 
 #include "common/alignment.h"
@@ -16,6 +18,7 @@
 
 namespace VideoCommon::Shader {
 
+using std::move;
 using Tegra::Shader::AtomicOp;
 using Tegra::Shader::AtomicType;
 using Tegra::Shader::Attribute;
@@ -27,29 +30,26 @@ using Tegra::Shader::StoreType;
 
 namespace {
 
-Node GetAtomOperation(AtomicOp op, bool is_signed, Node memory, Node data) {
-    const OperationCode operation_code = [op] {
-        switch (op) {
-        case AtomicOp::Add:
-            return OperationCode::AtomicIAdd;
-        case AtomicOp::Min:
-            return OperationCode::AtomicIMin;
-        case AtomicOp::Max:
-            return OperationCode::AtomicIMax;
-        case AtomicOp::And:
-            return OperationCode::AtomicIAnd;
-        case AtomicOp::Or:
-            return OperationCode::AtomicIOr;
-        case AtomicOp::Xor:
-            return OperationCode::AtomicIXor;
-        case AtomicOp::Exch:
-            return OperationCode::AtomicIExchange;
-        default:
-            UNIMPLEMENTED_MSG("op={}", static_cast<int>(op));
-            return OperationCode::AtomicIAdd;
-        }
-    }();
-    return SignedOperation(operation_code, is_signed, std::move(memory), std::move(data));
+OperationCode GetAtomOperation(AtomicOp op) {
+    switch (op) {
+    case AtomicOp::Add:
+        return OperationCode::AtomicIAdd;
+    case AtomicOp::Min:
+        return OperationCode::AtomicIMin;
+    case AtomicOp::Max:
+        return OperationCode::AtomicIMax;
+    case AtomicOp::And:
+        return OperationCode::AtomicIAnd;
+    case AtomicOp::Or:
+        return OperationCode::AtomicIOr;
+    case AtomicOp::Xor:
+        return OperationCode::AtomicIXor;
+    case AtomicOp::Exch:
+        return OperationCode::AtomicIExchange;
+    default:
+        UNIMPLEMENTED_MSG("op={}", static_cast<int>(op));
+        return OperationCode::AtomicIAdd;
+    }
 }
 
 bool IsUnaligned(Tegra::Shader::UniformType uniform_type) {
@@ -90,23 +90,22 @@ u32 GetMemorySize(Tegra::Shader::UniformType uniform_type) {
 
 Node ExtractUnaligned(Node value, Node address, u32 mask, u32 size) {
     Node offset = Operation(OperationCode::UBitwiseAnd, address, Immediate(mask));
-    offset = Operation(OperationCode::ULogicalShiftLeft, std::move(offset), Immediate(3));
-    return Operation(OperationCode::UBitfieldExtract, std::move(value), std::move(offset),
-                     Immediate(size));
+    offset = Operation(OperationCode::ULogicalShiftLeft, move(offset), Immediate(3));
+    return Operation(OperationCode::UBitfieldExtract, move(value), move(offset), Immediate(size));
 }
 
 Node InsertUnaligned(Node dest, Node value, Node address, u32 mask, u32 size) {
-    Node offset = Operation(OperationCode::UBitwiseAnd, std::move(address), Immediate(mask));
-    offset = Operation(OperationCode::ULogicalShiftLeft, std::move(offset), Immediate(3));
-    return Operation(OperationCode::UBitfieldInsert, std::move(dest), std::move(value),
-                     std::move(offset), Immediate(size));
+    Node offset = Operation(OperationCode::UBitwiseAnd, move(address), Immediate(mask));
+    offset = Operation(OperationCode::ULogicalShiftLeft, move(offset), Immediate(3));
+    return Operation(OperationCode::UBitfieldInsert, move(dest), move(value), move(offset),
+                     Immediate(size));
 }
 
 Node Sign16Extend(Node value) {
     Node sign = Operation(OperationCode::UBitwiseAnd, value, Immediate(1U << 15));
-    Node is_sign = Operation(OperationCode::LogicalUEqual, std::move(sign), Immediate(1U << 15));
+    Node is_sign = Operation(OperationCode::LogicalUEqual, move(sign), Immediate(1U << 15));
     Node extend = Operation(OperationCode::Select, is_sign, Immediate(0xFFFF0000), Immediate(0));
-    return Operation(OperationCode::UBitwiseOr, std::move(value), std::move(extend));
+    return Operation(OperationCode::UBitwiseOr, move(value), move(extend));
 }
 
 } // Anonymous namespace
@@ -379,20 +378,36 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
 
             if (IsUnaligned(type)) {
                 const u32 mask = GetUnalignedMask(type);
-                value = InsertUnaligned(gmem, std::move(value), real_address, mask, size);
+                value = InsertUnaligned(gmem, move(value), real_address, mask, size);
             }
 
             bb.push_back(Operation(OperationCode::Assign, gmem, value));
         }
         break;
     }
+    case OpCode::Id::RED: {
+        UNIMPLEMENTED_IF_MSG(instr.red.type != GlobalAtomicType::U32);
+        UNIMPLEMENTED_IF_MSG(instr.red.operation != AtomicOp::Add);
+        const auto [real_address, base_address, descriptor] =
+            TrackGlobalMemory(bb, instr, true, true);
+        if (!real_address || !base_address) {
+            // Tracking failed, skip atomic.
+            break;
+        }
+        Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
+        Node value = GetRegister(instr.gpr0);
+        bb.push_back(Operation(OperationCode::ReduceIAdd, move(gmem), move(value)));
+        break;
+    }
     case OpCode::Id::ATOM: {
         UNIMPLEMENTED_IF_MSG(instr.atom.operation == AtomicOp::Inc ||
                                  instr.atom.operation == AtomicOp::Dec ||
                                  instr.atom.operation == AtomicOp::SafeAdd,
                              "operation={}", static_cast<int>(instr.atom.operation.Value()));
         UNIMPLEMENTED_IF_MSG(instr.atom.type == GlobalAtomicType::S64 ||
-                                 instr.atom.type == GlobalAtomicType::U64,
+                                 instr.atom.type == GlobalAtomicType::U64 ||
+                                 instr.atom.type == GlobalAtomicType::F16x2_FTZ_RN ||
+                                 instr.atom.type == GlobalAtomicType::F32_FTZ_RN,
                              "type={}", static_cast<int>(instr.atom.type.Value()));
 
         const auto [real_address, base_address, descriptor] =
@@ -403,11 +418,11 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }
 
         const bool is_signed =
-            instr.atoms.type == AtomicType::S32 || instr.atoms.type == AtomicType::S64;
+            instr.atom.type == GlobalAtomicType::S32 || instr.atom.type == GlobalAtomicType::S64;
         Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
-        Node value = GetAtomOperation(static_cast<AtomicOp>(instr.atom.operation), is_signed, gmem,
-                                      GetRegister(instr.gpr20));
-        SetRegister(bb, instr.gpr0, std::move(value));
+        SetRegister(bb, instr.gpr0,
+                    SignedOperation(GetAtomOperation(instr.atom.operation), is_signed, gmem,
+                                    GetRegister(instr.gpr20)));
         break;
     }
     case OpCode::Id::ATOMS: {
@@ -421,11 +436,10 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             instr.atoms.type == AtomicType::S32 || instr.atoms.type == AtomicType::S64;
         const s32 offset = instr.atoms.GetImmediateOffset();
         Node address = GetRegister(instr.gpr8);
-        address = Operation(OperationCode::IAdd, std::move(address), Immediate(offset));
-        Node value =
-            GetAtomOperation(static_cast<AtomicOp>(instr.atoms.operation), is_signed,
-                             GetSharedMemory(std::move(address)), GetRegister(instr.gpr20));
-        SetRegister(bb, instr.gpr0, std::move(value));
+        address = Operation(OperationCode::IAdd, move(address), Immediate(offset));
+        SetRegister(bb, instr.gpr0,
+                    SignedOperation(GetAtomOperation(instr.atoms.operation), is_signed,
+                                    GetSharedMemory(move(address)), GetRegister(instr.gpr20)));
         break;
     }
     case OpCode::Id::AL2P: {

src/video_core/shader/node.h

@@ -178,6 +178,20 @@ enum class OperationCode {
     AtomicIOr,       /// (memory, int) -> int
     AtomicIXor,      /// (memory, int) -> int
 
+    ReduceUAdd, /// (memory, uint) -> void
+    ReduceUMin, /// (memory, uint) -> void
+    ReduceUMax, /// (memory, uint) -> void
+    ReduceUAnd, /// (memory, uint) -> void
+    ReduceUOr,  /// (memory, uint) -> void
+    ReduceUXor, /// (memory, uint) -> void
+
+    ReduceIAdd, /// (memory, int) -> void
+    ReduceIMin, /// (memory, int) -> void
+    ReduceIMax, /// (memory, int) -> void
+    ReduceIAnd, /// (memory, int) -> void
+    ReduceIOr,  /// (memory, int) -> void
+    ReduceIXor, /// (memory, int) -> void
+
     Branch,         /// (uint branch_target) -> void
     BranchIndirect, /// (uint branch_target) -> void
     PushFlowStack,  /// (uint branch_target) -> void

src/video_core/texture_cache/texture_cache.h

@@ -509,7 +509,9 @@ private:
         }
         const auto& final_params = new_surface->GetSurfaceParams();
         if (cr_params.type != final_params.type) {
-            BufferCopy(current_surface, new_surface);
+            if (Settings::values.use_accurate_gpu_emulation) {
+                BufferCopy(current_surface, new_surface);
+            }
         } else {
             std::vector<CopyParams> bricks = current_surface->BreakDown(final_params);
             for (auto& brick : bricks) {