decode/arithmetic_half: Fix HADD2 and HMUL2 absolute and negation bits

The encoding for negation and absolute value was wrong.
Extracting is now done manually. Similar instructions having different
encodings is the rule, not the exception. To keep sanity and readability
I preferred to extract the desired bit manually.

This is implemented against nxas:
https://github.com/ReinUsesLisp/nxas/blob/8dbc38995711cc12206aa370145a3a02665fd989/table.h#L68

That is itself tested against nvdisasm (Nvidia's official disassembler).

CMakeLists.txt

@@ -164,7 +164,7 @@ if (ENABLE_SDL2)
             set(SDL2_LIBRARIES "SDL2::SDL2")
         endif()
 
-        include_directories(${SDL2_INCLUDE_DIRS})
+        include_directories(SYSTEM ${SDL2_INCLUDE_DIRS})
         add_library(SDL2 INTERFACE)
         target_link_libraries(SDL2 INTERFACE "${SDL2_LIBRARIES}")
     endif()

externals/CMakeLists.txt

@@ -16,11 +16,11 @@ add_library(catch-single-include INTERFACE)
 target_include_directories(catch-single-include INTERFACE catch/single_include)
 
 # libfmt
-pkg_check_modules(FMT IMPORTED_TARGET GLOBAL fmt>=6.1.0)
+pkg_check_modules(FMT IMPORTED_TARGET GLOBAL fmt>=6.2.0)
 if (FMT_FOUND)
     add_library(fmt::fmt ALIAS PkgConfig::FMT)
 else()
-    message(STATUS "fmt 6.1.0 or newer not found, falling back to externals")
+    message(STATUS "fmt 6.2.0 or newer not found, falling back to externals")
     add_subdirectory(fmt)
     add_library(fmt::fmt ALIAS fmt)
 endif()

src/common/file_util.cpp

@@ -888,7 +888,14 @@ std::string SanitizePath(std::string_view path_, DirectorySeparator directory_se
     }
 
     std::replace(path.begin(), path.end(), type1, type2);
-    path.erase(std::unique(path.begin(), path.end(),
+
+    auto start = path.begin();
+#ifdef _WIN32
+    // allow network paths which start with a double backslash (e.g. \\server\share)
+    if (start != path.end())
+        ++start;
+#endif
+    path.erase(std::unique(start, path.end(),
                            [type2](char c1, char c2) { return c1 == type2 && c2 == type2; }),
                path.end());
     return std::string(RemoveTrailingSlash(path));

src/core/hle/kernel/memory/slab_heap.h

@@ -51,7 +51,7 @@ public:
     }
 
     void Free(void* obj) {
-        Node* node = reinterpret_cast<Node*>(obj);
+        Node* node = static_cast<Node*>(obj);
 
         Node* cur_head = head.load();
         do {
@@ -145,7 +145,7 @@ public:
     }
 
     T* Allocate() {
-        T* obj = reinterpret_cast<T*>(AllocateImpl());
+        T* obj = static_cast<T*>(AllocateImpl());
         if (obj != nullptr) {
             new (obj) T();
         }

src/core/hle/service/hid/controllers/npad.cpp

@@ -107,6 +107,7 @@ void Controller_NPad::InitNewlyAddedControler(std::size_t controller_idx) {
     switch (controller_type) {
     case NPadControllerType::None:
         UNREACHABLE();
+        break;
     case NPadControllerType::Handheld:
         controller.joy_styles.handheld.Assign(1);
         controller.device_type.handheld.Assign(1);
@@ -363,6 +364,7 @@ void Controller_NPad::OnUpdate(const Core::Timing::CoreTiming& core_timing, u8*
         switch (controller_type) {
         case NPadControllerType::None:
             UNREACHABLE();
+            break;
         case NPadControllerType::Handheld:
             handheld_entry.connection_status.raw = 0;
             handheld_entry.connection_status.IsWired.Assign(1);

src/core/hle/service/time/standard_network_system_clock_core.h

@@ -23,7 +23,7 @@ public:
         standard_network_clock_sufficient_accuracy = value;
     }
 
-    bool IsStandardNetworkSystemClockAccuracySufficient(Core::System& system) {
+    bool IsStandardNetworkSystemClockAccuracySufficient(Core::System& system) const {
         SystemClockContext context{};
         if (GetClockContext(system, context) != RESULT_SUCCESS) {
             return {};

src/core/hle/service/time/steady_clock_core.h

@@ -16,6 +16,7 @@ namespace Service::Time::Clock {
 class SteadyClockCore {
 public:
     SteadyClockCore() = default;
+    virtual ~SteadyClockCore() = default;
 
     const Common::UUID& GetClockSourceId() const {
         return clock_source_id;

src/core/hle/service/time/system_clock_context_update_callback.h

@@ -20,7 +20,7 @@ namespace Service::Time::Clock {
 class SystemClockContextUpdateCallback {
 public:
     SystemClockContextUpdateCallback();
-    ~SystemClockContextUpdateCallback();
+    virtual ~SystemClockContextUpdateCallback();
 
     bool NeedUpdate(const SystemClockContext& value) const;
 

src/core/hle/service/time/system_clock_core.cpp

@@ -9,7 +9,7 @@
 namespace Service::Time::Clock {
 
 SystemClockCore::SystemClockCore(SteadyClockCore& steady_clock_core)
-    : steady_clock_core{steady_clock_core}, is_initialized{} {
+    : steady_clock_core{steady_clock_core} {
     context.steady_time_point.clock_source_id = steady_clock_core.GetClockSourceId();
 }
 

src/core/hle/service/time/system_clock_core.h

@@ -22,7 +22,7 @@ class SystemClockContextUpdateCallback;
 class SystemClockCore {
 public:
     explicit SystemClockCore(SteadyClockCore& steady_clock_core);
-    ~SystemClockCore();
+    virtual ~SystemClockCore();
 
     SteadyClockCore& GetSteadyClockCore() const {
         return steady_clock_core;

src/core/loader/nso.cpp

@@ -37,7 +37,7 @@ static_assert(sizeof(MODHeader) == 0x1c, "MODHeader has incorrect size.");
 
 std::vector<u8> DecompressSegment(const std::vector<u8>& compressed_data,
                                   const NSOSegmentHeader& header) {
-    const std::vector<u8> uncompressed_data =
+    std::vector<u8> uncompressed_data =
         Common::Compression::DecompressDataLZ4(compressed_data, header.size);
 
     ASSERT_MSG(uncompressed_data.size() == header.size, "{} != {}", header.size,

src/video_core/engines/shader_bytecode.h

@@ -817,11 +817,9 @@ union Instruction {
         BitField<32, 1, u64> saturate;
         BitField<49, 2, HalfMerge> merge;
 
-        BitField<43, 1, u64> negate_a;
         BitField<44, 1, u64> abs_a;
         BitField<47, 2, HalfType> type_a;
 
-        BitField<31, 1, u64> negate_b;
         BitField<30, 1, u64> abs_b;
         BitField<28, 2, HalfType> type_b;
 

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -1145,6 +1145,7 @@ private:
                 return {"gl_FragCoord"s + GetSwizzle(element), Type::Float};
             default:
                 UNREACHABLE();
+                return {"0", Type::Int};
             }
         case Attribute::Index::FrontColor:
             return {"gl_Color"s + GetSwizzle(element), Type::Float};

src/video_core/shader/decode/arithmetic_half.cpp

@@ -19,22 +19,46 @@ u32 ShaderIR::DecodeArithmeticHalf(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
 
-    if (opcode->get().GetId() == OpCode::Id::HADD2_C ||
-        opcode->get().GetId() == OpCode::Id::HADD2_R) {
+    bool negate_a = false;
+    bool negate_b = false;
+    bool absolute_a = false;
+    bool absolute_b = false;
+
+    switch (opcode->get().GetId()) {
+    case OpCode::Id::HADD2_R:
         if (instr.alu_half.ftz == 0) {
             LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
         }
+        negate_a = ((instr.value >> 43) & 1) != 0;
+        negate_b = ((instr.value >> 31) & 1) != 0;
+        absolute_a = ((instr.value >> 44) & 1) != 0;
+        absolute_b = ((instr.value >> 30) & 1) != 0;
+        break;
+    case OpCode::Id::HADD2_C:
+        if (instr.alu_half.ftz == 0) {
+            LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
+        }
+        negate_a = ((instr.value >> 43) & 1) != 0;
+        negate_b = ((instr.value >> 56) & 1) != 0;
+        absolute_a = ((instr.value >> 44) & 1) != 0;
+        absolute_b = ((instr.value >> 54) & 1) != 0;
+        break;
+    case OpCode::Id::HMUL2_R:
+        negate_a = ((instr.value >> 43) & 1) != 0;
+        absolute_a = ((instr.value >> 44) & 1) != 0;
+        absolute_b = ((instr.value >> 30) & 1) != 0;
+        break;
+    case OpCode::Id::HMUL2_C:
+        negate_b = ((instr.value >> 31) & 1) != 0;
+        absolute_a = ((instr.value >> 44) & 1) != 0;
+        absolute_b = ((instr.value >> 54) & 1) != 0;
+        break;
     }
 
-    const bool negate_a =
-        opcode->get().GetId() != OpCode::Id::HMUL2_R && instr.alu_half.negate_a != 0;
-    const bool negate_b =
-        opcode->get().GetId() != OpCode::Id::HMUL2_C && instr.alu_half.negate_b != 0;
-
     Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half.type_a);
-    op_a = GetOperandAbsNegHalf(op_a, instr.alu_half.abs_a, negate_a);
+    op_a = GetOperandAbsNegHalf(op_a, absolute_a, negate_a);
 
-    auto [type_b, op_b] = [&]() -> std::tuple<HalfType, Node> {
+    auto [type_b, op_b] = [this, instr, opcode]() -> std::pair<HalfType, Node> {
         switch (opcode->get().GetId()) {
         case OpCode::Id::HADD2_C:
         case OpCode::Id::HMUL2_C:
@@ -48,17 +72,16 @@ u32 ShaderIR::DecodeArithmeticHalf(NodeBlock& bb, u32 pc) {
         }
     }();
     op_b = UnpackHalfFloat(op_b, type_b);
-    // redeclaration to avoid a bug in clang with reusing local bindings in lambdas
-    Node op_b_alt = GetOperandAbsNegHalf(op_b, instr.alu_half.abs_b, negate_b);
+    op_b = GetOperandAbsNegHalf(op_b, absolute_b, negate_b);
 
-    Node value = [&]() {
+    Node value = [this, opcode, op_a, op_b = op_b] {
         switch (opcode->get().GetId()) {
         case OpCode::Id::HADD2_C:
         case OpCode::Id::HADD2_R:
-            return Operation(OperationCode::HAdd, PRECISE, op_a, op_b_alt);
+            return Operation(OperationCode::HAdd, PRECISE, op_a, op_b);
         case OpCode::Id::HMUL2_C:
         case OpCode::Id::HMUL2_R:
-            return Operation(OperationCode::HMul, PRECISE, op_a, op_b_alt);
+            return Operation(OperationCode::HMul, PRECISE, op_a, op_b);
         default:
             UNIMPLEMENTED_MSG("Unhandled half float instruction: {}", opcode->get().GetName());
             return Immediate(0);