shader/memory: Implement ATOM.ADD

ATOM operates atomically on global memory. For now only add ATOM.ADD
since that's what was found in commercial games.

This asserts for ATOM.ADD.S32 (handling the others as unimplemented),
although ATOM.ADD.U32 shouldn't be any different.

This change forces us to change the default type on SPIR-V storage
buffers from float to uint. We could also alias the buffers, but it's
simpler for now to just use uint. While we are at it, abstract the code
to avoid repetition.

src/common/logging/backend.cpp

@@ -120,7 +120,7 @@ private:
             duration_cast<std::chrono::microseconds>(steady_clock::now() - time_origin);
         entry.log_class = log_class;
         entry.log_level = log_level;
-        entry.filename = Common::TrimSourcePath(filename);
+        entry.filename = filename;
         entry.line_num = line_nr;
         entry.function = function;
         entry.message = std::move(message);

src/common/logging/backend.h

@@ -23,7 +23,7 @@ struct Entry {
     std::chrono::microseconds timestamp;
     Class log_class;
     Level log_level;
-    std::string filename;
+    const char* filename;
     unsigned int line_num;
     std::string function;
     std::string message;

src/common/logging/log.h

@@ -9,6 +9,15 @@
 
 namespace Log {
 
+// trims up to and including the last of ../, ..\, src/, src\ in a string
+constexpr const char* TrimSourcePath(std::string_view source) {
+    const auto rfind = [source](const std::string_view match) {
+        return source.rfind(match) == source.npos ? 0 : (source.rfind(match) + match.size());
+    };
+    auto idx = std::max({rfind("src/"), rfind("src\\"), rfind("../"), rfind("..\\")});
+    return source.data() + idx;
+}
+
 /// Specifies the severity or level of detail of the log message.
 enum class Level : u8 {
     Trace,    ///< Extremely detailed and repetitive debugging information that is likely to
@@ -141,24 +150,24 @@ void FmtLogMessage(Class log_class, Level log_level, const char* filename, unsig
 
 #ifdef _DEBUG
 #define LOG_TRACE(log_class, ...)                                                                  \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Trace, __FILE__, __LINE__,         \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Trace,                             \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #else
 #define LOG_TRACE(log_class, fmt, ...) (void(0))
 #endif
 
 #define LOG_DEBUG(log_class, ...)                                                                  \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Debug, __FILE__, __LINE__,         \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Debug,                             \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #define LOG_INFO(log_class, ...)                                                                   \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Info, __FILE__, __LINE__,          \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Info,                              \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #define LOG_WARNING(log_class, ...)                                                                \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Warning, __FILE__, __LINE__,       \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Warning,                           \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #define LOG_ERROR(log_class, ...)                                                                  \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Error, __FILE__, __LINE__,         \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Error,                             \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #define LOG_CRITICAL(log_class, ...)                                                               \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Critical, __FILE__, __LINE__,      \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Critical,                          \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)

src/common/string_util.cpp

@@ -223,26 +223,4 @@ std::u16string UTF16StringFromFixedZeroTerminatedBuffer(std::u16string_view buff
     return std::u16string(buffer.begin(), buffer.begin() + len);
 }
 
-const char* TrimSourcePath(const char* path, const char* root) {
-    const char* p = path;
-
-    while (*p != '\0') {
-        const char* next_slash = p;
-        while (*next_slash != '\0' && *next_slash != '/' && *next_slash != '\\') {
-            ++next_slash;
-        }
-
-        bool is_src = Common::ComparePartialString(p, next_slash, root);
-        p = next_slash;
-
-        if (*p != '\0') {
-            ++p;
-        }
-        if (is_src) {
-            path = p;
-        }
-    }
-    return path;
-}
-
 } // namespace Common

src/core/loader/nso.cpp

@@ -97,7 +97,8 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::Process& process,
         if (nso_header.IsSegmentCompressed(i)) {
             data = DecompressSegment(data, nso_header.segments[i]);
         }
-        program_image.resize(nso_header.segments[i].location + data.size());
+        program_image.resize(nso_header.segments[i].location +
+                             PageAlignSize(static_cast<u32>(data.size())));
         std::memcpy(program_image.data() + nso_header.segments[i].location, data.data(),
                     data.size());
         codeset.segments[i].addr = nso_header.segments[i].location;
@@ -105,8 +106,12 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::Process& process,
         codeset.segments[i].size = PageAlignSize(static_cast<u32>(data.size()));
     }
 
-    if (should_pass_arguments && !Settings::values.program_args.empty()) {
-        const auto arg_data = Settings::values.program_args;
+    if (should_pass_arguments) {
+        std::vector<u8> arg_data{Settings::values.program_args.begin(),
+                                 Settings::values.program_args.end()};
+        if (arg_data.empty()) {
+            arg_data.resize(NSO_ARGUMENT_DEFAULT_SIZE);
+        }
         codeset.DataSegment().size += NSO_ARGUMENT_DATA_ALLOCATION_SIZE;
         NSOArgumentHeader args_header{
             NSO_ARGUMENT_DATA_ALLOCATION_SIZE, static_cast<u32_le>(arg_data.size()), {}};

src/core/loader/nso.h

@@ -56,6 +56,8 @@ static_assert(sizeof(NSOHeader) == 0x100, "NSOHeader has incorrect size.");
 static_assert(std::is_trivially_copyable_v<NSOHeader>, "NSOHeader must be trivially copyable.");
 
 constexpr u64 NSO_ARGUMENT_DATA_ALLOCATION_SIZE = 0x9000;
+// NOTE: Official software default argument state is unverified.
+constexpr u64 NSO_ARGUMENT_DEFAULT_SIZE = 1;
 
 struct NSOArgumentHeader {
     u32_le allocated_size;

src/video_core/engines/shader_bytecode.h

@@ -227,6 +227,28 @@ enum class AtomicOp : u64 {
     Exch = 8,
 };
 
+enum class GlobalAtomicOp : u64 {
+    Add = 0,
+    Min = 1,
+    Max = 2,
+    Inc = 3,
+    Dec = 4,
+    And = 5,
+    Or = 6,
+    Xor = 7,
+    Exch = 8,
+    SafeAdd = 10,
+};
+
+enum class GlobalAtomicType : u64 {
+    U32 = 0,
+    S32 = 1,
+    U64 = 2,
+    F32_FTZ_RN = 3,
+    F16x2_FTZ_RN = 4,
+    S64 = 5,
+};
+
 enum class UniformType : u64 {
     UnsignedByte = 0,
     SignedByte = 1,
@@ -957,6 +979,12 @@ union Instruction {
         BitField<46, 2, u64> cache_mode;
     } stg;
 
+    union {
+        BitField<52, 4, GlobalAtomicOp> operation;
+        BitField<49, 3, GlobalAtomicType> type;
+        BitField<28, 20, s64> offset;
+    } atom;
+
     union {
         BitField<52, 4, AtomicOp> operation;
         BitField<28, 2, AtomicType> type;
@@ -1690,6 +1718,7 @@ public:
         ST_S,
         ST,    // Store in generic memory
         STG,   // Store in global memory
+        ATOM,  // Atomic operation on global memory
         ATOMS, // Atomic operation on shared memory
         AL2P,  // Transforms attribute memory into physical memory
         TEX,
@@ -1994,6 +2023,7 @@ 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("11101101--------", Id::ATOM, Type::Memory, "ATOM"),
             INST("11101100--------", Id::ATOMS, Type::Memory, "ATOMS"),
             INST("1110111110100---", Id::AL2P, Type::Memory, "AL2P"),
             INST("110000----111---", Id::TEX, Type::Texture, "TEX"),

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -1858,10 +1858,7 @@ private:
 
     template <const std::string_view& opname, Type type>
     Expression Atomic(Operation operation) {
-        ASSERT(stage == ShaderType::Compute);
-        auto& smem = std::get<SmemNode>(*operation[0]);
-
-        return {fmt::format("atomic{}(smem[{} >> 2], {})", opname, Visit(smem.GetAddress()).AsInt(),
+        return {fmt::format("atomic{}({}, {})", opname, Visit(operation[0]).GetCode(),
                             Visit(operation[1]).As(type)),
                 type};
     }

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -325,9 +325,6 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
     specialization.tessellation.primitive = fixed_state.tessellation.primitive;
     specialization.tessellation.spacing = fixed_state.tessellation.spacing;
     specialization.tessellation.clockwise = fixed_state.tessellation.clockwise;
-    for (const auto& rt : key.renderpass_params.color_attachments) {
-        specialization.enabled_rendertargets.set(rt.index);
-    }
 
     SPIRVProgram program;
     std::vector<vk::DescriptorSetLayoutBinding> bindings;

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -542,11 +542,10 @@ private:
             return;
         }
 
-        for (u32 rt = 0; rt < static_cast<u32>(frag_colors.size()); ++rt) {
-            if (!specialization.enabled_rendertargets[rt]) {
+        for (u32 rt = 0; rt < static_cast<u32>(std::size(frag_colors)); ++rt) {
+            if (!IsRenderTargetEnabled(rt)) {
                 continue;
             }
-
             const Id id = AddGlobalVariable(OpVariable(t_out_float4, spv::StorageClass::Output));
             Name(id, fmt::format("frag_color{}", rt));
             Decorate(id, spv::Decoration::Location, rt);
@@ -852,6 +851,15 @@ private:
         return binding;
     }
 
+    bool IsRenderTargetEnabled(u32 rt) const {
+        for (u32 component = 0; component < 4; ++component) {
+            if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
+                return true;
+            }
+        }
+        return false;
+    }
+
     bool IsInputAttributeArray() const {
         return stage == ShaderType::TesselationControl || stage == ShaderType::TesselationEval ||
                stage == ShaderType::Geometry;
@@ -1115,15 +1123,7 @@ private:
         }
 
         if (const auto gmem = std::get_if<GmemNode>(&*node)) {
-            const Id gmem_buffer = global_buffers.at(gmem->GetDescriptor());
-            const Id real = AsUint(Visit(gmem->GetRealAddress()));
-            const Id base = AsUint(Visit(gmem->GetBaseAddress()));
-
-            Id offset = OpISub(t_uint, real, base);
-            offset = OpUDiv(t_uint, offset, Constant(t_uint, 4U));
-            return {OpLoad(t_float,
-                           OpAccessChain(t_gmem_float, gmem_buffer, Constant(t_uint, 0U), offset)),
-                    Type::Float};
+            return {OpLoad(t_uint, GetGlobalMemoryPointer(*gmem)), Type::Uint};
         }
 
         if (const auto lmem = std::get_if<LmemNode>(&*node)) {
@@ -1134,10 +1134,7 @@ private:
         }
 
         if (const auto smem = std::get_if<SmemNode>(&*node)) {
-            Id address = AsUint(Visit(smem->GetAddress()));
-            address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
-            const Id pointer = OpAccessChain(t_smem_uint, shared_memory, address);
-            return {OpLoad(t_uint, pointer), Type::Uint};
+            return {OpLoad(t_uint, GetSharedMemoryPointer(*smem)), Type::Uint};
         }
 
         if (const auto internal_flag = std::get_if<InternalFlagNode>(&*node)) {
@@ -1331,20 +1328,10 @@ private:
             target = {OpAccessChain(t_prv_float, local_memory, address), Type::Float};
 
         } else if (const auto smem = std::get_if<SmemNode>(&*dest)) {
-            ASSERT(stage == ShaderType::Compute);
-            Id address = AsUint(Visit(smem->GetAddress()));
-            address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
-            target = {OpAccessChain(t_smem_uint, shared_memory, address), Type::Uint};
+            target = {GetSharedMemoryPointer(*smem), Type::Uint};
 
         } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) {
-            const Id real = AsUint(Visit(gmem->GetRealAddress()));
-            const Id base = AsUint(Visit(gmem->GetBaseAddress()));
-            const Id diff = OpISub(t_uint, real, base);
-            const Id offset = OpShiftRightLogical(t_uint, diff, Constant(t_uint, 2));
-
-            const Id gmem_buffer = global_buffers.at(gmem->GetDescriptor());
-            target = {OpAccessChain(t_gmem_float, gmem_buffer, Constant(t_uint, 0), offset),
-                      Type::Float};
+            target = {GetGlobalMemoryPointer(*gmem), Type::Uint};
 
         } else {
             UNIMPLEMENTED();
@@ -1796,11 +1783,16 @@ private:
         return {};
     }
 
-    Expression UAtomicAdd(Operation operation) {
-        const auto& smem = std::get<SmemNode>(*operation[0]);
-        Id address = AsUint(Visit(smem.GetAddress()));
-        address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
-        const Id pointer = OpAccessChain(t_smem_uint, shared_memory, address);
+    Expression AtomicAdd(Operation operation) {
+        Id pointer;
+        if (const auto smem = std::get_if<SmemNode>(&*operation[0])) {
+            pointer = GetSharedMemoryPointer(*smem);
+        } else if (const auto gmem = std::get_if<GmemNode>(&*operation[0])) {
+            pointer = GetGlobalMemoryPointer(*gmem);
+        } else {
+            UNREACHABLE();
+            return {Constant(t_uint, 0), Type::Uint};
+        }
 
         const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
         const Id semantics = Constant(t_uint, 0U);
@@ -1889,19 +1881,14 @@ private:
             // rendertargets/components are skipped in the register assignment.
             u32 current_reg = 0;
             for (u32 rt = 0; rt < Maxwell::NumRenderTargets; ++rt) {
-                if (!specialization.enabled_rendertargets[rt]) {
-                    // Skip rendertargets that are not enabled
-                    continue;
-                }
                 // TODO(Subv): Figure out how dual-source blending is configured in the Switch.
                 for (u32 component = 0; component < 4; ++component) {
-                    const Id pointer = AccessElement(t_out_float, frag_colors.at(rt), component);
-                    if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
-                        OpStore(pointer, SafeGetRegister(current_reg));
-                        ++current_reg;
-                    } else {
-                        OpStore(pointer, component == 3 ? v_float_one : v_float_zero);
+                    if (!header.ps.IsColorComponentOutputEnabled(rt, component)) {
+                        continue;
                     }
+                    const Id pointer = AccessElement(t_out_float, frag_colors[rt], component);
+                    OpStore(pointer, SafeGetRegister(current_reg));
+                    ++current_reg;
                 }
             }
             if (header.ps.omap.depth) {
@@ -2240,6 +2227,22 @@ private:
         return {};
     }
 
+    Id GetGlobalMemoryPointer(const GmemNode& gmem) {
+        const Id real = AsUint(Visit(gmem.GetRealAddress()));
+        const Id base = AsUint(Visit(gmem.GetBaseAddress()));
+        const Id diff = OpISub(t_uint, real, base);
+        const Id offset = OpShiftRightLogical(t_uint, diff, Constant(t_uint, 2));
+        const Id buffer = global_buffers.at(gmem.GetDescriptor());
+        return OpAccessChain(t_gmem_uint, buffer, Constant(t_uint, 0), offset);
+    }
+
+    Id GetSharedMemoryPointer(const SmemNode& smem) {
+        ASSERT(stage == ShaderType::Compute);
+        Id address = AsUint(Visit(smem.GetAddress()));
+        address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
+        return OpAccessChain(t_smem_uint, shared_memory, address);
+    }
+
     static constexpr std::array operation_decompilers = {
         &SPIRVDecompiler::Assign,
 
@@ -2386,7 +2389,7 @@ private:
         &SPIRVDecompiler::AtomicImageXor,
         &SPIRVDecompiler::AtomicImageExchange,
 
-        &SPIRVDecompiler::UAtomicAdd,
+        &SPIRVDecompiler::AtomicAdd,
 
         &SPIRVDecompiler::Branch,
         &SPIRVDecompiler::BranchIndirect,
@@ -2482,9 +2485,9 @@ private:
 
     Id t_smem_uint{};
 
-    const Id t_gmem_float = TypePointer(spv::StorageClass::StorageBuffer, t_float);
+    const Id t_gmem_uint = TypePointer(spv::StorageClass::StorageBuffer, t_uint);
     const Id t_gmem_array =
-        Name(Decorate(TypeRuntimeArray(t_float), spv::Decoration::ArrayStride, 4U), "GmemArray");
+        Name(Decorate(TypeRuntimeArray(t_uint), spv::Decoration::ArrayStride, 4U), "GmemArray");
     const Id t_gmem_struct = MemberDecorate(
         Decorate(TypeStruct(t_gmem_array), spv::Decoration::Block), 0, spv::Decoration::Offset, 0);
     const Id t_gmem_ssbo = TypePointer(spv::StorageClass::StorageBuffer, t_gmem_struct);

src/video_core/renderer_vulkan/vk_shader_decompiler.h

@@ -102,9 +102,6 @@ struct Specialization final {
         Maxwell::TessellationSpacing spacing{};
         bool clockwise{};
     } tessellation;
-
-    // Fragment specific
-    std::bitset<8> enabled_rendertargets;
 };
 // Old gcc versions don't consider this trivially copyable.
 // static_assert(std::is_trivially_copyable_v<Specialization>);

src/video_core/shader/decode/memory.cpp

@@ -19,6 +19,8 @@ namespace VideoCommon::Shader {
 using Tegra::Shader::AtomicOp;
 using Tegra::Shader::AtomicType;
 using Tegra::Shader::Attribute;
+using Tegra::Shader::GlobalAtomicOp;
+using Tegra::Shader::GlobalAtomicType;
 using Tegra::Shader::Instruction;
 using Tegra::Shader::OpCode;
 using Tegra::Shader::Register;
@@ -335,6 +337,24 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }
         break;
     }
+    case OpCode::Id::ATOM: {
+        UNIMPLEMENTED_IF_MSG(instr.atom.operation != GlobalAtomicOp::Add, "operation={}",
+                             static_cast<int>(instr.atom.operation.Value()));
+        UNIMPLEMENTED_IF_MSG(instr.atom.type != GlobalAtomicType::S32, "type={}",
+                             static_cast<int>(instr.atom.type.Value()));
+
+        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 = Operation(OperationCode::AtomicAdd, std::move(gmem), GetRegister(instr.gpr20));
+        SetRegister(bb, instr.gpr0, std::move(value));
+        break;
+    }
     case OpCode::Id::ATOMS: {
         UNIMPLEMENTED_IF_MSG(instr.atoms.operation != AtomicOp::Add, "operation={}",
                              static_cast<int>(instr.atoms.operation.Value()));
@@ -348,7 +368,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         Node memory = GetSharedMemory(std::move(address));
         Node data = GetRegister(instr.gpr20);
 
-        Node value = Operation(OperationCode::UAtomicAdd, std::move(memory), std::move(data));
+        Node value = Operation(OperationCode::AtomicAdd, std::move(memory), std::move(data));
         SetRegister(bb, instr.gpr0, std::move(value));
         break;
     }

src/video_core/shader/decode/texture.cpp

@@ -161,16 +161,16 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
     case OpCode::Id::TXD: {
         UNIMPLEMENTED_IF_MSG(instr.txd.UsesMiscMode(TextureMiscMode::AOFFI),
                              "AOFFI is not implemented");
-        UNIMPLEMENTED_IF_MSG(instr.txd.is_array != 0, "TXD Array is not implemented");
 
+        const bool is_array = instr.txd.is_array != 0;
         u64 base_reg = instr.gpr8.Value();
         const auto derivate_reg = instr.gpr20.Value();
         const auto texture_type = instr.txd.texture_type.Value();
         const auto coord_count = GetCoordCount(texture_type);
 
-        const Sampler* sampler = is_bindless
-                                     ? GetBindlessSampler(base_reg, {{texture_type, false, false}})
-                                     : GetSampler(instr.sampler, {{texture_type, false, false}});
+        const Sampler* sampler =
+            is_bindless ? GetBindlessSampler(base_reg, {{texture_type, is_array, false}})
+                        : GetSampler(instr.sampler, {{texture_type, is_array, false}});
         Node4 values;
         if (sampler == nullptr) {
             for (u32 element = 0; element < values.size(); ++element) {
@@ -179,6 +179,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
             WriteTexInstructionFloat(bb, instr, values);
             break;
         }
+
         if (is_bindless) {
             base_reg++;
         }
@@ -192,8 +193,14 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
             derivates.push_back(GetRegister(derivate_reg + derivate + 1));
         }
 
+        Node array_node = {};
+        if (is_array) {
+            const Node info_reg = GetRegister(base_reg + coord_count);
+            array_node = BitfieldExtract(info_reg, 0, 16);
+        }
+
         for (u32 element = 0; element < values.size(); ++element) {
-            MetaTexture meta{*sampler, {}, {}, {}, {}, derivates, {}, {}, {}, element};
+            MetaTexture meta{*sampler, array_node, {}, {}, {}, derivates, {}, {}, {}, element};
             values[element] = Operation(OperationCode::TextureGradient, std::move(meta), coords);
         }
 

src/video_core/shader/node.h

@@ -162,7 +162,7 @@ enum class OperationCode {
     AtomicImageXor,      /// (MetaImage, int[N] coords) -> void
     AtomicImageExchange, /// (MetaImage, int[N] coords) -> void
 
-    UAtomicAdd, /// (smem, uint) -> uint
+    AtomicAdd, /// (memory, {u}int) -> {u}int
 
     Branch,         /// (uint branch_target) -> void
     BranchIndirect, /// (uint branch_target) -> void

src/yuzu/CMakeLists.txt

@@ -36,9 +36,6 @@ add_executable(yuzu
     configuration/configure_filesystem.cpp
     configuration/configure_filesystem.h
     configuration/configure_filesystem.ui
-    configuration/configure_gamelist.cpp
-    configuration/configure_gamelist.h
-    configuration/configure_gamelist.ui
     configuration/configure_general.cpp
     configuration/configure_general.h
     configuration/configure_general.ui
@@ -75,6 +72,9 @@ add_executable(yuzu
     configuration/configure_touchscreen_advanced.cpp
     configuration/configure_touchscreen_advanced.h
     configuration/configure_touchscreen_advanced.ui
+    configuration/configure_ui.cpp
+    configuration/configure_ui.h
+    configuration/configure_ui.ui
     configuration/configure_web.cpp
     configuration/configure_web.h
     configuration/configure_web.ui

src/yuzu/configuration/configure.ui

@@ -48,7 +48,7 @@
          <string>General</string>
         </attribute>
        </widget>
-       <widget class="ConfigureGameList" name="gameListTab">
+       <widget class="ConfigureUi" name="uiTab">
         <attribute name="title">
          <string>Game List</string>
         </attribute>
@@ -166,9 +166,9 @@
    <container>1</container>
   </customwidget>
   <customwidget>
-   <class>ConfigureGameList</class>
+   <class>ConfigureUi</class>
    <extends>QWidget</extends>
-   <header>configuration/configure_gamelist.h</header>
+   <header>configuration/configure_ui.h</header>
    <container>1</container>
   </customwidget>
   <customwidget>

src/yuzu/configuration/configure_dialog.cpp

@@ -34,7 +34,7 @@ void ConfigureDialog::SetConfiguration() {}
 
 void ConfigureDialog::ApplyConfiguration() {
     ui->generalTab->ApplyConfiguration();
-    ui->gameListTab->ApplyConfiguration();
+    ui->uiTab->ApplyConfiguration();
     ui->systemTab->ApplyConfiguration();
     ui->profileManagerTab->ApplyConfiguration();
     ui->filesystemTab->applyConfiguration();
@@ -74,7 +74,7 @@ Q_DECLARE_METATYPE(QList<QWidget*>);
 
 void ConfigureDialog::PopulateSelectionList() {
     const std::array<std::pair<QString, QList<QWidget*>>, 5> items{
-        {{tr("General"), {ui->generalTab, ui->webTab, ui->debugTab, ui->gameListTab}},
+        {{tr("General"), {ui->generalTab, ui->webTab, ui->debugTab, ui->uiTab}},
          {tr("System"), {ui->systemTab, ui->profileManagerTab, ui->serviceTab, ui->filesystemTab}},
          {tr("Graphics"), {ui->graphicsTab}},
          {tr("Audio"), {ui->audioTab}},
@@ -108,7 +108,7 @@ void ConfigureDialog::UpdateVisibleTabs() {
         {ui->audioTab, tr("Audio")},
         {ui->debugTab, tr("Debug")},
         {ui->webTab, tr("Web")},
-        {ui->gameListTab, tr("Game List")},
+        {ui->uiTab, tr("UI")},
         {ui->filesystemTab, tr("Filesystem")},
         {ui->serviceTab, tr("Services")},
     };

src/yuzu/configuration/configure_general.cpp

@@ -15,11 +15,6 @@ ConfigureGeneral::ConfigureGeneral(QWidget* parent)
 
     ui->setupUi(this);
 
-    for (const auto& theme : UISettings::themes) {
-        ui->theme_combobox->addItem(QString::fromUtf8(theme.first),
-                                    QString::fromUtf8(theme.second));
-    }
-
     SetConfiguration();
 
     connect(ui->toggle_frame_limit, &QCheckBox::toggled, ui->frame_limit, &QSpinBox::setEnabled);
@@ -30,7 +25,6 @@ ConfigureGeneral::~ConfigureGeneral() = default;
 void ConfigureGeneral::SetConfiguration() {
     ui->toggle_check_exit->setChecked(UISettings::values.confirm_before_closing);
     ui->toggle_user_on_boot->setChecked(UISettings::values.select_user_on_boot);
-    ui->theme_combobox->setCurrentIndex(ui->theme_combobox->findData(UISettings::values.theme));
     ui->toggle_background_pause->setChecked(UISettings::values.pause_when_in_background);
 
     ui->toggle_frame_limit->setChecked(Settings::values.use_frame_limit);
@@ -41,8 +35,6 @@ void ConfigureGeneral::SetConfiguration() {
 void ConfigureGeneral::ApplyConfiguration() {
     UISettings::values.confirm_before_closing = ui->toggle_check_exit->isChecked();
     UISettings::values.select_user_on_boot = ui->toggle_user_on_boot->isChecked();
-    UISettings::values.theme =
-        ui->theme_combobox->itemData(ui->theme_combobox->currentIndex()).toString();
     UISettings::values.pause_when_in_background = ui->toggle_background_pause->isChecked();
 
     Settings::values.use_frame_limit = ui->toggle_frame_limit->isChecked();

src/yuzu/configuration/configure_general.ui

@@ -65,39 +65,12 @@
             </property>
            </widget>
           </item>
-           <item>
-             <widget class="QCheckBox" name="toggle_background_pause">
-               <property name="text">
-                 <string>Pause emulation when in background</string>
-               </property>
-             </widget>
-           </item>
-         </layout>
-        </item>
-       </layout>
-      </widget>
-     </item>
-     <item>
-      <widget class="QGroupBox" name="theme_group_box">
-       <property name="title">
-        <string>Theme</string>
-       </property>
-       <layout class="QHBoxLayout" name="theme_qhbox_layout">
-        <item>
-         <layout class="QVBoxLayout" name="theme_qvbox_layout">
           <item>
-           <layout class="QHBoxLayout" name="theme_qhbox_layout_2">
-            <item>
-             <widget class="QLabel" name="theme_label">
-              <property name="text">
-               <string>Theme:</string>
-              </property>
-             </widget>
-            </item>
-            <item>
-             <widget class="QComboBox" name="theme_combobox"/>
-            </item>
-           </layout>
+           <widget class="QCheckBox" name="toggle_background_pause">
+            <property name="text">
+             <string>Pause emulation when in background</string>
+            </property>
+           </widget>
           </item>
          </layout>
         </item>

src/yuzu/configuration/configure_ui.cpp

@@ -7,8 +7,8 @@
 
 #include "common/common_types.h"
 #include "core/settings.h"
-#include "ui_configure_gamelist.h"
-#include "yuzu/configuration/configure_gamelist.h"
+#include "ui_configure_ui.h"
+#include "yuzu/configuration/configure_ui.h"
 #include "yuzu/uisettings.h"
 
 namespace {
@@ -26,35 +26,40 @@ constexpr std::array row_text_names{
 };
 } // Anonymous namespace
 
-ConfigureGameList::ConfigureGameList(QWidget* parent)
-    : QWidget(parent), ui(new Ui::ConfigureGameList) {
+ConfigureUi::ConfigureUi(QWidget* parent) : QWidget(parent), ui(new Ui::ConfigureUi) {
     ui->setupUi(this);
 
+    for (const auto& theme : UISettings::themes) {
+        ui->theme_combobox->addItem(QString::fromUtf8(theme.first),
+                                    QString::fromUtf8(theme.second));
+    }
+
     InitializeIconSizeComboBox();
     InitializeRowComboBoxes();
 
     SetConfiguration();
 
     // Force game list reload if any of the relevant settings are changed.
-    connect(ui->show_unknown, &QCheckBox::stateChanged, this,
-            &ConfigureGameList::RequestGameListUpdate);
+    connect(ui->show_unknown, &QCheckBox::stateChanged, this, &ConfigureUi::RequestGameListUpdate);
     connect(ui->icon_size_combobox, QOverload<int>::of(&QComboBox::currentIndexChanged), this,
-            &ConfigureGameList::RequestGameListUpdate);
+            &ConfigureUi::RequestGameListUpdate);
     connect(ui->row_1_text_combobox, QOverload<int>::of(&QComboBox::currentIndexChanged), this,
-            &ConfigureGameList::RequestGameListUpdate);
+            &ConfigureUi::RequestGameListUpdate);
     connect(ui->row_2_text_combobox, QOverload<int>::of(&QComboBox::currentIndexChanged), this,
-            &ConfigureGameList::RequestGameListUpdate);
+            &ConfigureUi::RequestGameListUpdate);
 
     // Update text ComboBoxes after user interaction.
     connect(ui->row_1_text_combobox, QOverload<int>::of(&QComboBox::activated),
-            [=]() { ConfigureGameList::UpdateSecondRowComboBox(); });
+            [=]() { ConfigureUi::UpdateSecondRowComboBox(); });
     connect(ui->row_2_text_combobox, QOverload<int>::of(&QComboBox::activated),
-            [=]() { ConfigureGameList::UpdateFirstRowComboBox(); });
+            [=]() { ConfigureUi::UpdateFirstRowComboBox(); });
 }
 
-ConfigureGameList::~ConfigureGameList() = default;
+ConfigureUi::~ConfigureUi() = default;
 
-void ConfigureGameList::ApplyConfiguration() {
+void ConfigureUi::ApplyConfiguration() {
+    UISettings::values.theme =
+        ui->theme_combobox->itemData(ui->theme_combobox->currentIndex()).toString();
     UISettings::values.show_unknown = ui->show_unknown->isChecked();
     UISettings::values.show_add_ons = ui->show_add_ons->isChecked();
     UISettings::values.icon_size = ui->icon_size_combobox->currentData().toUInt();
@@ -63,18 +68,19 @@ void ConfigureGameList::ApplyConfiguration() {
     Settings::Apply();
 }
 
-void ConfigureGameList::RequestGameListUpdate() {
+void ConfigureUi::RequestGameListUpdate() {
     UISettings::values.is_game_list_reload_pending.exchange(true);
 }
 
-void ConfigureGameList::SetConfiguration() {
+void ConfigureUi::SetConfiguration() {
+    ui->theme_combobox->setCurrentIndex(ui->theme_combobox->findData(UISettings::values.theme));
     ui->show_unknown->setChecked(UISettings::values.show_unknown);
     ui->show_add_ons->setChecked(UISettings::values.show_add_ons);
     ui->icon_size_combobox->setCurrentIndex(
         ui->icon_size_combobox->findData(UISettings::values.icon_size));
 }
 
-void ConfigureGameList::changeEvent(QEvent* event) {
+void ConfigureUi::changeEvent(QEvent* event) {
     if (event->type() == QEvent::LanguageChange) {
         RetranslateUI();
     }
@@ -82,7 +88,7 @@ void ConfigureGameList::changeEvent(QEvent* event) {
     QWidget::changeEvent(event);
 }
 
-void ConfigureGameList::RetranslateUI() {
+void ConfigureUi::RetranslateUI() {
     ui->retranslateUi(this);
 
     for (int i = 0; i < ui->icon_size_combobox->count(); i++) {
@@ -97,18 +103,18 @@ void ConfigureGameList::RetranslateUI() {
     }
 }
 
-void ConfigureGameList::InitializeIconSizeComboBox() {
+void ConfigureUi::InitializeIconSizeComboBox() {
     for (const auto& size : default_icon_sizes) {
         ui->icon_size_combobox->addItem(QString::fromUtf8(size.second), size.first);
     }
 }
 
-void ConfigureGameList::InitializeRowComboBoxes() {
+void ConfigureUi::InitializeRowComboBoxes() {
     UpdateFirstRowComboBox(true);
     UpdateSecondRowComboBox(true);
 }
 
-void ConfigureGameList::UpdateFirstRowComboBox(bool init) {
+void ConfigureUi::UpdateFirstRowComboBox(bool init) {
     const int currentIndex =
         init ? UISettings::values.row_1_text_id
              : ui->row_1_text_combobox->findData(ui->row_1_text_combobox->currentData());
@@ -127,7 +133,7 @@ void ConfigureGameList::UpdateFirstRowComboBox(bool init) {
         ui->row_1_text_combobox->findData(ui->row_2_text_combobox->currentData()));
 }
 
-void ConfigureGameList::UpdateSecondRowComboBox(bool init) {
+void ConfigureUi::UpdateSecondRowComboBox(bool init) {
     const int currentIndex =
         init ? UISettings::values.row_2_text_id
              : ui->row_2_text_combobox->findData(ui->row_2_text_combobox->currentData());

src/yuzu/configuration/configure_ui.h

@@ -8,15 +8,15 @@
 #include <QWidget>
 
 namespace Ui {
-class ConfigureGameList;
+class ConfigureUi;
 }
 
-class ConfigureGameList : public QWidget {
+class ConfigureUi : public QWidget {
     Q_OBJECT
 
 public:
-    explicit ConfigureGameList(QWidget* parent = nullptr);
-    ~ConfigureGameList() override;
+    explicit ConfigureUi(QWidget* parent = nullptr);
+    ~ConfigureUi() override;
 
     void ApplyConfiguration();
 
@@ -34,5 +34,5 @@ private:
     void UpdateFirstRowComboBox(bool init = false);
     void UpdateSecondRowComboBox(bool init = false);
 
-    std::unique_ptr<Ui::ConfigureGameList> ui;
+    std::unique_ptr<Ui::ConfigureUi> ui;
 };

src/yuzu/configuration/configure_ui.ui

@@ -1,7 +1,7 @@
 <?xml version="1.0" encoding="UTF-8"?>
 <ui version="4.0">
- <class>ConfigureGameList</class>
- <widget class="QWidget" name="ConfigureGameList">
+ <class>ConfigureUi</class>
+ <widget class="QWidget" name="ConfigureUi">
   <property name="geometry">
    <rect>
     <x>0</x>
@@ -21,7 +21,34 @@
        <property name="title">
         <string>General</string>
        </property>
-       <layout class="QHBoxLayout" name="GeneralHorizontalLayout">
+       <layout class="QHBoxLayout" name="horizontalLayout">
+        <item>
+         <layout class="QVBoxLayout" name="verticalLayout">
+          <item>
+           <layout class="QHBoxLayout" name="horizontalLayout_3">
+            <item>
+             <widget class="QLabel" name="theme_label">
+              <property name="text">
+               <string>Theme:</string>
+              </property>
+             </widget>
+            </item>
+            <item>
+             <widget class="QComboBox" name="theme_combobox"/>
+            </item>
+           </layout>
+          </item>
+         </layout>
+        </item>
+       </layout>
+      </widget>
+     </item>
+     <item>
+      <widget class="QGroupBox" name="GameListGroupBox">
+       <property name="title">
+        <string>Game List</string>
+       </property>
+       <layout class="QHBoxLayout" name="GameListHorizontalLayout">
         <item>
          <layout class="QVBoxLayout" name="GeneralVerticalLayout">
           <item>
@@ -38,19 +65,6 @@
             </property>
            </widget>
           </item>
-         </layout>
-        </item>
-       </layout>
-      </widget>
-     </item>
-     <item>
-      <widget class="QGroupBox" name="IconSizeGroupBox">
-       <property name="title">
-        <string>Icon Size</string>
-       </property>
-       <layout class="QHBoxLayout" name="icon_size_qhbox_layout">
-        <item>
-         <layout class="QVBoxLayout" name="icon_size_qvbox_layout">
           <item>
            <layout class="QHBoxLayout" name="icon_size_qhbox_layout_2">
             <item>
@@ -65,19 +79,6 @@
             </item>
            </layout>
           </item>
-         </layout>
-        </item>
-       </layout>
-      </widget>
-     </item>
-     <item>
-      <widget class="QGroupBox" name="RowGroupBox">
-       <property name="title">
-        <string>Row Text</string>
-       </property>
-       <layout class="QHBoxLayout" name="RowHorizontalLayout">
-        <item>
-         <layout class="QVBoxLayout" name="RowVerticalLayout">
           <item>
            <layout class="QHBoxLayout" name="row_1_qhbox_layout">
             <item>