Merge pull request #3506 from namkazt/patch-9

shader_decode: Implement partial ATOM/ATOMS instr

src/video_core/engines/shader_bytecode.h

@@ -231,18 +231,6 @@ enum class AtomicOp : u64 {
     Or = 6,
     Xor = 7,
     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,
 };
 
@@ -1001,7 +989,7 @@ union Instruction {
     } stg;
 
     union {
-        BitField<52, 4, GlobalAtomicOp> operation;
+        BitField<52, 4, AtomicOp> operation;
         BitField<49, 3, GlobalAtomicType> type;
         BitField<28, 20, s64> offset;
     } atom;

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -2114,6 +2114,10 @@ private:
 
     template <const std::string_view& opname, Type type>
     Expression Atomic(Operation operation) {
+        if ((opname == Func::Min || opname == Func::Max) && type == Type::Int) {
+            UNIMPLEMENTED_MSG("Unimplemented Min & Max for atomic operations");
+            return {};
+        }
         return {fmt::format("atomic{}({}, {})", opname, Visit(operation[0]).GetCode(),
                             Visit(operation[1]).As(type)),
                 type};
@@ -2307,6 +2311,8 @@ private:
         ~Func() = delete;
 
         static constexpr std::string_view Add = "Add";
+        static constexpr std::string_view Min = "Min";
+        static constexpr std::string_view Max = "Max";
         static constexpr std::string_view And = "And";
         static constexpr std::string_view Or = "Or";
         static constexpr std::string_view Xor = "Xor";
@@ -2457,7 +2463,21 @@ private:
         &GLSLDecompiler::AtomicImage<Func::Xor>,
         &GLSLDecompiler::AtomicImage<Func::Exchange>,
 
+        &GLSLDecompiler::Atomic<Func::Exchange, Type::Uint>,
         &GLSLDecompiler::Atomic<Func::Add, Type::Uint>,
+        &GLSLDecompiler::Atomic<Func::Min, Type::Uint>,
+        &GLSLDecompiler::Atomic<Func::Max, Type::Uint>,
+        &GLSLDecompiler::Atomic<Func::And, Type::Uint>,
+        &GLSLDecompiler::Atomic<Func::Or, Type::Uint>,
+        &GLSLDecompiler::Atomic<Func::Xor, Type::Uint>,
+
+        &GLSLDecompiler::Atomic<Func::Exchange, Type::Int>,
+        &GLSLDecompiler::Atomic<Func::Add, Type::Int>,
+        &GLSLDecompiler::Atomic<Func::Min, Type::Int>,
+        &GLSLDecompiler::Atomic<Func::Max, Type::Int>,
+        &GLSLDecompiler::Atomic<Func::And, Type::Int>,
+        &GLSLDecompiler::Atomic<Func::Or, Type::Int>,
+        &GLSLDecompiler::Atomic<Func::Xor, Type::Int>,
 
         &GLSLDecompiler::Branch,
         &GLSLDecompiler::BranchIndirect,

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -1941,7 +1941,11 @@ private:
         return {};
     }
 
-    Expression AtomicAdd(Operation operation) {
+    template <Id (Module::*func)(Id, Id, Id, Id, Id), Type result_type,
+              Type value_type = result_type>
+    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);
@@ -1949,14 +1953,15 @@ private:
             pointer = GetGlobalMemoryPointer(*gmem);
         } else {
             UNREACHABLE();
-            return {Constant(t_uint, 0), Type::Uint};
+            return {Constant(type_def, 0), result_type};
         }
 
-        const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
+        const Id value = As(Visit(operation[1]), value_type);
-        const Id semantics = Constant(t_uint, 0U);
 
-        const Id value = AsUint(Visit(operation[1]));
+        const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
-        return {OpAtomicIAdd(t_uint, pointer, scope, semantics, value), Type::Uint};
+        const Id semantics = Constant(type_def, 0);
+
+        return {(this->*func)(type_def, pointer, scope, semantics, value), result_type};
     }
 
     Expression Branch(Operation operation) {
@@ -2545,7 +2550,21 @@ private:
         &SPIRVDecompiler::AtomicImageXor,
         &SPIRVDecompiler::AtomicImageExchange,
 
-        &SPIRVDecompiler::AtomicAdd,
+        &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, 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::Branch,
         &SPIRVDecompiler::BranchIndirect,

src/video_core/shader/decode/memory.cpp

@@ -19,7 +19,6 @@ 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;
@@ -28,6 +27,28 @@ 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;
+        }
+    }();
+    return SignedOperation(operation_code, is_signed, std::move(memory), std::move(data));
+}
+
 bool IsUnaligned(Tegra::Shader::UniformType uniform_type) {
     return uniform_type == Tegra::Shader::UniformType::UnsignedByte ||
            uniform_type == Tegra::Shader::UniformType::UnsignedShort;
@@ -363,10 +384,13 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         break;
     }
     case OpCode::Id::ATOM: {
-        UNIMPLEMENTED_IF_MSG(instr.atom.operation != GlobalAtomicOp::Add, "operation={}",
+        UNIMPLEMENTED_IF_MSG(instr.atom.operation == AtomicOp::Inc ||
-                             static_cast<int>(instr.atom.operation.Value()));
+                                 instr.atom.operation == AtomicOp::Dec ||
-        UNIMPLEMENTED_IF_MSG(instr.atom.type != GlobalAtomicType::S32, "type={}",
+                                 instr.atom.operation == AtomicOp::SafeAdd,
-                             static_cast<int>(instr.atom.type.Value()));
+                             "operation={}", static_cast<int>(instr.atom.operation.Value()));
+        UNIMPLEMENTED_IF_MSG(instr.atom.type == GlobalAtomicType::S64 ||
+                                 instr.atom.type == GlobalAtomicType::U64,
+                             "type={}", static_cast<int>(instr.atom.type.Value()));
 
         const auto [real_address, base_address, descriptor] =
             TrackGlobalMemory(bb, instr, true, true);
@@ -375,25 +399,29 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             break;
         }
 
+        const bool is_signed =
+            instr.atoms.type == AtomicType::S32 || instr.atoms.type == AtomicType::S64;
         Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
-        Node value = Operation(OperationCode::AtomicAdd, std::move(gmem), GetRegister(instr.gpr20));
+        Node value = GetAtomOperation(static_cast<AtomicOp>(instr.atom.operation), is_signed, 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={}",
+        UNIMPLEMENTED_IF_MSG(instr.atoms.operation == AtomicOp::Inc ||
-                             static_cast<int>(instr.atoms.operation.Value()));
+                                 instr.atoms.operation == AtomicOp::Dec,
-        UNIMPLEMENTED_IF_MSG(instr.atoms.type != AtomicType::U32, "type={}",
+                             "operation={}", static_cast<int>(instr.atoms.operation.Value()));
-                             static_cast<int>(instr.atoms.type.Value()));
+        UNIMPLEMENTED_IF_MSG(instr.atoms.type == AtomicType::S64 ||
-
+                                 instr.atoms.type == AtomicType::U64,
+                             "type={}", static_cast<int>(instr.atoms.type.Value()));
+        const bool is_signed =
+            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 =
-        Node memory = GetSharedMemory(std::move(address));
+            GetAtomOperation(static_cast<AtomicOp>(instr.atoms.operation), is_signed,
-        Node data = GetRegister(instr.gpr20);
+                             GetSharedMemory(std::move(address)), GetRegister(instr.gpr20));
-
-        Node value = Operation(OperationCode::AtomicAdd, std::move(memory), std::move(data));
         SetRegister(bb, instr.gpr0, std::move(value));
         break;
     }

src/video_core/shader/node.h

@@ -162,7 +162,21 @@ enum class OperationCode {
     AtomicImageXor,      /// (MetaImage, int[N] coords) -> void
     AtomicImageExchange, /// (MetaImage, int[N] coords) -> void
 
-    AtomicAdd, /// (memory, {u}int) -> {u}int
+    AtomicUExchange, /// (memory, uint) -> uint
+    AtomicUAdd,      /// (memory, uint) -> uint
+    AtomicUMin,      /// (memory, uint) -> uint
+    AtomicUMax,      /// (memory, uint) -> uint
+    AtomicUAnd,      /// (memory, uint) -> uint
+    AtomicUOr,       /// (memory, uint) -> uint
+    AtomicUXor,      /// (memory, uint) -> uint
+
+    AtomicIExchange, /// (memory, int) -> int
+    AtomicIAdd,      /// (memory, int) -> int
+    AtomicIMin,      /// (memory, int) -> int
+    AtomicIMax,      /// (memory, int) -> int
+    AtomicIAnd,      /// (memory, int) -> int
+    AtomicIOr,       /// (memory, int) -> int
+    AtomicIXor,      /// (memory, int) -> int
 
     Branch,         /// (uint branch_target) -> void
     BranchIndirect, /// (uint branch_target) -> void

src/video_core/shader/node_helper.cpp

@@ -86,6 +86,20 @@ OperationCode SignedToUnsignedCode(OperationCode operation_code, bool is_signed)
         return OperationCode::LogicalUNotEqual;
     case OperationCode::LogicalIGreaterEqual:
         return OperationCode::LogicalUGreaterEqual;
+    case OperationCode::AtomicIExchange:
+        return OperationCode::AtomicUExchange;
+    case OperationCode::AtomicIAdd:
+        return OperationCode::AtomicUAdd;
+    case OperationCode::AtomicIMin:
+        return OperationCode::AtomicUMin;
+    case OperationCode::AtomicIMax:
+        return OperationCode::AtomicUMax;
+    case OperationCode::AtomicIAnd:
+        return OperationCode::AtomicUAnd;
+    case OperationCode::AtomicIOr:
+        return OperationCode::AtomicUOr;
+    case OperationCode::AtomicIXor:
+        return OperationCode::AtomicUXor;
     case OperationCode::INegate:
         UNREACHABLE_MSG("Can't negate an unsigned integer");
         return {};