Merge pull request #4147 from ReinUsesLisp/hset2-imm

shader/half_set: Implement HSET2_IMM

src/video_core/engines/shader_bytecode.h

@@ -661,6 +661,10 @@ union Instruction {
     constexpr Instruction(u64 value) : value{value} {}
     constexpr Instruction(const Instruction& instr) : value(instr.value) {}
 
+    constexpr bool Bit(u64 offset) const {
+        return ((value >> offset) & 1) != 0;
+    }
+
     BitField<0, 8, Register> gpr0;
     BitField<8, 8, Register> gpr8;
     union {
@@ -1874,7 +1878,9 @@ public:
         HSETP2_C,
         HSETP2_R,
         HSETP2_IMM,
+        HSET2_C,
         HSET2_R,
+        HSET2_IMM,
         POPC_C,
         POPC_R,
         POPC_IMM,
@@ -2194,7 +2200,9 @@ private:
             INST("0111111-1-------", Id::HSETP2_C, Type::HalfSetPredicate, "HSETP2_C"),
             INST("0101110100100---", Id::HSETP2_R, Type::HalfSetPredicate, "HSETP2_R"),
             INST("0111111-0-------", Id::HSETP2_IMM, Type::HalfSetPredicate, "HSETP2_IMM"),
+            INST("0111110-1-------", Id::HSET2_C, Type::HalfSet, "HSET2_C"),
             INST("0101110100011---", Id::HSET2_R, Type::HalfSet, "HSET2_R"),
+            INST("0111110-0-------", Id::HSET2_IMM, Type::HalfSet, "HSET2_IMM"),
             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"),

src/video_core/shader/decode/half_set.cpp

@@ -13,55 +13,101 @@
 
 namespace VideoCommon::Shader {
 
+using std::move;
 using Tegra::Shader::Instruction;
 using Tegra::Shader::OpCode;
+using Tegra::Shader::PredCondition;
 
 u32 ShaderIR::DecodeHalfSet(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
 
-    if (instr.hset2.ftz == 0) {
-        LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
+    PredCondition cond;
+    bool bf;
+    bool ftz;
+    bool neg_a;
+    bool abs_a;
+    bool neg_b;
+    bool abs_b;
+    switch (opcode->get().GetId()) {
+    case OpCode::Id::HSET2_C:
+    case OpCode::Id::HSET2_IMM:
+        cond = instr.hsetp2.cbuf_and_imm.cond;
+        bf = instr.Bit(53);
+        ftz = instr.Bit(54);
+        neg_a = instr.Bit(43);
+        abs_a = instr.Bit(44);
+        neg_b = instr.Bit(56);
+        abs_b = instr.Bit(54);
+        break;
+    case OpCode::Id::HSET2_R:
+        cond = instr.hsetp2.reg.cond;
+        bf = instr.Bit(49);
+        ftz = instr.Bit(50);
+        neg_a = instr.Bit(43);
+        abs_a = instr.Bit(44);
+        neg_b = instr.Bit(31);
+        abs_b = instr.Bit(30);
+        break;
+    default:
+        UNREACHABLE();
     }
 
-    Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hset2.type_a);
-    op_a = GetOperandAbsNegHalf(op_a, instr.hset2.abs_a, instr.hset2.negate_a);
-
-    Node op_b = [&]() {
+    Node op_b = [this, instr, opcode] {
         switch (opcode->get().GetId()) {
+        case OpCode::Id::HSET2_C:
+            // Inform as unimplemented as this is not tested.
+            UNIMPLEMENTED_MSG("HSET2_C is not implemented");
+            return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
         case OpCode::Id::HSET2_R:
             return GetRegister(instr.gpr20);
+        case OpCode::Id::HSET2_IMM:
+            return UnpackHalfImmediate(instr, true);
         default:
             UNREACHABLE();
-            return Immediate(0);
+            return Node{};
         }
     }();
-    op_b = UnpackHalfFloat(op_b, instr.hset2.type_b);
-    op_b = GetOperandAbsNegHalf(op_b, instr.hset2.abs_b, instr.hset2.negate_b);
 
-    const Node second_pred = GetPredicate(instr.hset2.pred39, instr.hset2.neg_pred);
+    if (!ftz) {
+        LOG_DEBUG(HW_GPU, "{} without FTZ is not implemented", opcode->get().GetName());
+    }
+
+    Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hset2.type_a);
+    op_a = GetOperandAbsNegHalf(op_a, abs_a, neg_a);
+
+    switch (opcode->get().GetId()) {
+    case OpCode::Id::HSET2_R:
+        op_b = GetOperandAbsNegHalf(move(op_b), abs_b, neg_b);
+        [[fallthrough]];
+    case OpCode::Id::HSET2_C:
+        op_b = UnpackHalfFloat(move(op_b), instr.hset2.type_b);
+        break;
+    default:
+        break;
+    }
 
-    const Node comparison_pair = GetPredicateComparisonHalf(instr.hset2.cond, op_a, op_b);
+    Node second_pred = GetPredicate(instr.hset2.pred39, instr.hset2.neg_pred);
+
+    Node comparison_pair = GetPredicateComparisonHalf(cond, op_a, op_b);
 
     const OperationCode combiner = GetPredicateCombiner(instr.hset2.op);
 
     // HSET2 operates on each half float in the pack.
     std::array<Node, 2> values;
     for (u32 i = 0; i < 2; ++i) {
-        const u32 raw_value = instr.hset2.bf ? 0x3c00 : 0xffff;
-        const Node true_value = Immediate(raw_value << (i * 16));
-        const Node false_value = Immediate(0);
-
-        const Node comparison =
-            Operation(OperationCode::LogicalPick2, comparison_pair, Immediate(i));
-        const Node predicate = Operation(combiner, comparison, second_pred);
+        const u32 raw_value = bf ? 0x3c00 : 0xffff;
+        Node true_value = Immediate(raw_value << (i * 16));
+        Node false_value = Immediate(0);
 
+        Node comparison = Operation(OperationCode::LogicalPick2, comparison_pair, Immediate(i));
+        Node predicate = Operation(combiner, comparison, second_pred);
         values[i] =
-            Operation(OperationCode::Select, NO_PRECISE, predicate, true_value, false_value);
+            Operation(OperationCode::Select, predicate, move(true_value), move(false_value));
     }
 
-    const Node value = Operation(OperationCode::UBitwiseOr, NO_PRECISE, values[0], values[1]);
-    SetRegister(bb, instr.gpr0, value);
+    Node value = Operation(OperationCode::UBitwiseOr, values[0], values[1]);
+    SetRegister(bb, instr.gpr0, move(value));
 
     return pc;
 }