shader_ir/decode: Implement half float saturation

src/video_core/shader/decode/arithmetic_half_immediate.cpp

@@ -23,8 +23,6 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) {
     } else {
         UNIMPLEMENTED_IF(instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::None);
     }
-    UNIMPLEMENTED_IF_MSG(instr.alu_half_imm.saturate != 0,
-                         "Half float immediate saturation not implemented");
 
     Node op_a = GetRegister(instr.gpr8);
     op_a = GetOperandAbsNegHalf(op_a, instr.alu_half_imm.abs_a, instr.alu_half_imm.negate_a);
@@ -43,10 +41,10 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) {
             return Immediate(0);
         }
     }();
+
+    value = GetSaturatedHalfFloat(value, instr.alu_half_imm.saturate);
     value = HalfMerge(GetRegister(instr.gpr0), value, instr.alu_half_imm.merge);
-
     SetRegister(bb, instr.gpr0, value);
-
     return pc;
 }
 

src/video_core/shader/shader_ir.cpp

@@ -218,6 +218,15 @@ Node ShaderIR::GetOperandAbsNegHalf(Node value, bool absolute, bool negate) {
     return value;
 }
 
+Node ShaderIR::GetSaturatedHalfFloat(Node value, bool saturate) {
+    if (!saturate) {
+        return value;
+    }
+    const Node positive_zero = Immediate(std::copysignf(0, 1));
+    const Node positive_one = Immediate(1.0f);
+    return Operation(OperationCode::HClamp, HALF_NO_PRECISE, value, positive_zero, positive_one);
+}
+
 Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, Node op_b) {
     const std::unordered_map<PredCondition, OperationCode> PredicateComparisonTable = {
         {PredCondition::LessThan, OperationCode::LogicalFLessThan},

src/video_core/shader/shader_ir.h

@@ -114,6 +114,7 @@ enum class OperationCode {
     HFma,      /// (MetaHalfArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2
     HAbsolute, /// (f16vec2 a) -> f16vec2
     HNegate,   /// (f16vec2 a, bool first, bool second) -> f16vec2
+    HClamp,    /// (f16vec2 src, float min, float max) -> f16vec2
     HMergeF32, /// (f16vec2 src) -> float
     HMergeH0,  /// (f16vec2 dest, f16vec2 src) -> f16vec2
     HMergeH1,  /// (f16vec2 dest, f16vec2 src) -> f16vec2
@@ -716,6 +717,8 @@ private:
     Node HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge);
     /// Conditionally absolute/negated half float pair. Absolute is applied first
     Node GetOperandAbsNegHalf(Node value, bool absolute, bool negate);
+    /// Conditionally saturates a half float pair
+    Node GetSaturatedHalfFloat(Node value, bool saturate = true);
 
     /// Returns a predicate comparing two floats
     Node GetPredicateComparisonFloat(Tegra::Shader::PredCondition condition, Node op_a, Node op_b);