Merge pull request #3799 from ReinUsesLisp/iadd-cc

shader: Implement P2R CC, IADD Rd.CC and IADD.X

externals/sirit

@@ -1 +1 @@
-Subproject commit a712959f1e373a33b48042b5934e288a243d5954
+Subproject commit 414fc4dbd28d8fe48f735a0c389db8a234f733c0

src/video_core/engines/shader_bytecode.h

@@ -813,6 +813,10 @@ union Instruction {
         BitField<49, 1, u64> negate_a;
     } alu_integer;
 
+    union {
+        BitField<43, 1, u64> x;
+    } iadd;
+
     union {
         BitField<39, 1, u64> ftz;
         BitField<32, 1, u64> saturate;

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -1870,6 +1870,14 @@ private:
         return GenerateBinaryInfix(operation, ">=", Type::Bool, type, type);
     }
 
+    Expression LogicalAddCarry(Operation operation) {
+        const std::string carry = code.GenerateTemporary();
+        code.AddLine("uint {};", carry);
+        code.AddLine("uaddCarry({}, {}, {});", VisitOperand(operation, 0).AsUint(),
+                     VisitOperand(operation, 1).AsUint(), carry);
+        return {fmt::format("({} != 0)", carry), Type::Bool};
+    }
+
     Expression LogicalFIsNan(Operation operation) {
         return GenerateUnary(operation, "isnan", Type::Bool, Type::Float);
     }
@@ -2441,6 +2449,8 @@ private:
         &GLSLDecompiler::LogicalNotEqual<Type::Uint>,
         &GLSLDecompiler::LogicalGreaterEqual<Type::Uint>,
 
+        &GLSLDecompiler::LogicalAddCarry,
+
         &GLSLDecompiler::Logical2HLessThan<false>,
         &GLSLDecompiler::Logical2HEqual<false>,
         &GLSLDecompiler::Logical2HLessEqual<false>,

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -1584,6 +1584,15 @@ private:
         return {OpCompositeConstruct(t_half, low, high), Type::HalfFloat};
     }
 
+    Expression LogicalAddCarry(Operation operation) {
+        const Id op_a = AsUint(Visit(operation[0]));
+        const Id op_b = AsUint(Visit(operation[1]));
+
+        const Id result = OpIAddCarry(TypeStruct({t_uint, t_uint}), op_a, op_b);
+        const Id carry = OpCompositeExtract(t_uint, result, 1);
+        return {OpINotEqual(t_bool, carry, Constant(t_uint, 0)), Type::Bool};
+    }
+
     Expression LogicalAssign(Operation operation) {
         const Node& dest = operation[0];
         const Node& src = operation[1];
@@ -2518,6 +2527,8 @@ private:
         &SPIRVDecompiler::Binary<&Module::OpINotEqual, Type::Bool, Type::Uint>,
         &SPIRVDecompiler::Binary<&Module::OpUGreaterThanEqual, Type::Bool, Type::Uint>,
 
+        &SPIRVDecompiler::LogicalAddCarry,
+
         &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool2, Type::HalfFloat>,
         &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool2, Type::HalfFloat>,
         &SPIRVDecompiler::Binary<&Module::OpFOrdLessThanEqual, Type::Bool2, Type::HalfFloat>,

src/video_core/shader/decode/arithmetic_integer.cpp

@@ -35,15 +35,38 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
     case OpCode::Id::IADD_C:
     case OpCode::Id::IADD_R:
     case OpCode::Id::IADD_IMM: {
-        UNIMPLEMENTED_IF_MSG(instr.alu.saturate_d, "IADD saturation not implemented");
+        UNIMPLEMENTED_IF_MSG(instr.alu.saturate_d, "IADD.SAT");
+        UNIMPLEMENTED_IF_MSG(instr.iadd.x && instr.generates_cc, "IADD.X Rd.CC");
 
         op_a = GetOperandAbsNegInteger(op_a, false, instr.alu_integer.negate_a, true);
         op_b = GetOperandAbsNegInteger(op_b, false, instr.alu_integer.negate_b, true);
 
-        const Node value = Operation(OperationCode::IAdd, PRECISE, op_a, op_b);
+        Node value = Operation(OperationCode::UAdd, op_a, op_b);
 
-        SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
+        if (instr.iadd.x) {
-        SetRegister(bb, instr.gpr0, value);
+            Node carry = GetInternalFlag(InternalFlag::Carry);
+            Node x = Operation(OperationCode::Select, std::move(carry), Immediate(1), Immediate(0));
+            value = Operation(OperationCode::UAdd, std::move(value), std::move(x));
+        }
+
+        if (instr.generates_cc) {
+            const Node i0 = Immediate(0);
+
+            Node zero = Operation(OperationCode::LogicalIEqual, value, i0);
+            Node sign = Operation(OperationCode::LogicalILessThan, value, i0);
+            Node carry = Operation(OperationCode::LogicalAddCarry, op_a, op_b);
+
+            Node pos_a = Operation(OperationCode::LogicalIGreaterThan, op_a, i0);
+            Node pos_b = Operation(OperationCode::LogicalIGreaterThan, op_b, i0);
+            Node pos = Operation(OperationCode::LogicalAnd, std::move(pos_a), std::move(pos_b));
+            Node overflow = Operation(OperationCode::LogicalAnd, pos, sign);
+
+            SetInternalFlag(bb, InternalFlag::Zero, std::move(zero));
+            SetInternalFlag(bb, InternalFlag::Sign, std::move(sign));
+            SetInternalFlag(bb, InternalFlag::Carry, std::move(carry));
+            SetInternalFlag(bb, InternalFlag::Overflow, std::move(overflow));
+        }
+        SetRegister(bb, instr.gpr0, std::move(value));
         break;
     }
     case OpCode::Id::IADD3_C:

src/video_core/shader/decode/register_set_predicate.cpp

@@ -2,6 +2,8 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <utility>
+
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "video_core/engines/shader_bytecode.h"
@@ -10,20 +12,20 @@
 
 namespace VideoCommon::Shader {
 
+using std::move;
 using Tegra::Shader::Instruction;
 using Tegra::Shader::OpCode;
 
 namespace {
-constexpr u64 NUM_PROGRAMMABLE_PREDICATES = 7;
+constexpr u64 NUM_CONDITION_CODES = 4;
-}
+constexpr u64 NUM_PREDICATES = 7;
+} // namespace
 
 u32 ShaderIR::DecodeRegisterSetPredicate(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
 
-    UNIMPLEMENTED_IF(instr.p2r_r2p.mode != Tegra::Shader::R2pMode::Pr);
+    Node apply_mask = [this, opcode, instr] {
-
-    const Node apply_mask = [&] {
         switch (opcode->get().GetId()) {
         case OpCode::Id::R2P_IMM:
         case OpCode::Id::P2R_IMM:
@@ -34,39 +36,43 @@ u32 ShaderIR::DecodeRegisterSetPredicate(NodeBlock& bb, u32 pc) {
         }
     }();
 
-    const auto offset = static_cast<u32>(instr.p2r_r2p.byte) * 8;
+    const u32 offset = static_cast<u32>(instr.p2r_r2p.byte) * 8;
+
+    const bool cc = instr.p2r_r2p.mode == Tegra::Shader::R2pMode::Cc;
+    const u64 num_entries = cc ? NUM_CONDITION_CODES : NUM_PREDICATES;
+    const auto get_entry = [this, cc](u64 entry) {
+        return cc ? GetInternalFlag(static_cast<InternalFlag>(entry)) : GetPredicate(entry);
+    };
 
     switch (opcode->get().GetId()) {
     case OpCode::Id::R2P_IMM: {
-        const Node mask = GetRegister(instr.gpr8);
+        Node mask = GetRegister(instr.gpr8);
 
-        for (u64 pred = 0; pred < NUM_PROGRAMMABLE_PREDICATES; ++pred) {
+        for (u64 entry = 0; entry < num_entries; ++entry) {
-            const auto shift = static_cast<u32>(pred);
+            const u32 shift = static_cast<u32>(entry);
 
-            const Node apply_compare = BitfieldExtract(apply_mask, shift, 1);
+            Node apply = BitfieldExtract(apply_mask, shift, 1);
-            const Node condition =
+            Node condition = Operation(OperationCode::LogicalUNotEqual, apply, Immediate(0));
-                Operation(OperationCode::LogicalUNotEqual, apply_compare, Immediate(0));
 
-            const Node value_compare = BitfieldExtract(mask, offset + shift, 1);
+            Node compare = BitfieldExtract(mask, offset + shift, 1);
-            const Node value =
+            Node value = Operation(OperationCode::LogicalUNotEqual, move(compare), Immediate(0));
-                Operation(OperationCode::LogicalUNotEqual, value_compare, Immediate(0));
 
-            const Node code = Operation(OperationCode::LogicalAssign, GetPredicate(pred), value);
+            Node code = Operation(OperationCode::LogicalAssign, get_entry(entry), move(value));
-            bb.push_back(Conditional(condition, {code}));
+            bb.push_back(Conditional(condition, {move(code)}));
         }
         break;
     }
     case OpCode::Id::P2R_IMM: {
         Node value = Immediate(0);
-        for (u64 pred = 0; pred < NUM_PROGRAMMABLE_PREDICATES; ++pred) {
+        for (u64 entry = 0; entry < num_entries; ++entry) {
-            Node bit = Operation(OperationCode::Select, GetPredicate(pred), Immediate(1U << pred),
+            Node bit = Operation(OperationCode::Select, get_entry(entry), Immediate(1U << entry),
                                  Immediate(0));
-            value = Operation(OperationCode::UBitwiseOr, std::move(value), std::move(bit));
+            value = Operation(OperationCode::UBitwiseOr, move(value), move(bit));
         }
-        value = Operation(OperationCode::UBitwiseAnd, std::move(value), apply_mask);
+        value = Operation(OperationCode::UBitwiseAnd, move(value), apply_mask);
-        value = BitfieldInsert(GetRegister(instr.gpr8), std::move(value), offset, 8);
+        value = BitfieldInsert(GetRegister(instr.gpr8), move(value), offset, 8);
 
-        SetRegister(bb, instr.gpr0, std::move(value));
+        SetRegister(bb, instr.gpr0, move(value));
         break;
     }
     default:

src/video_core/shader/node.h

@@ -132,6 +132,8 @@ enum class OperationCode {
     LogicalUNotEqual,     /// (uint a, uint b) -> bool
     LogicalUGreaterEqual, /// (uint a, uint b) -> bool
 
+    LogicalAddCarry, /// (uint a, uint b) -> bool
+
     Logical2HLessThan,            /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
     Logical2HEqual,               /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2
     Logical2HLessEqual,           /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2