Parser: Implement extension GL_AMD_gpu_shader_half_float.

- Add built-in types: float16_t, f16vec, f16mat.
- Add support of half float constant: hf, HF.
- Extend built-in floating-point operators: +, -, *, /, ++, --, +=, -=,
  *=, /=, ==, !=, >=, <=, >, <.
- Add support of type conversions: float16_t -> XXX, XXX -> float16_t.
- Add new built-in functions.

glslang/MachineIndependent/ParseHelper.cpp

@@ -4559,6 +4559,11 @@ void TParseContext::layoutTypeCheck(const TSourceLoc& loc, const TType& type)
         // containing a double, the offset must also be a multiple of 8..."
         if (type.containsBasicType(EbtDouble) && ! IsMultipleOfPow2(qualifier.layoutXfbOffset, 8))
             error(loc, "type contains double; xfb_offset must be a multiple of 8", "xfb_offset", "");
+#ifdef AMD_EXTENSIONS
+        // ..., if applied to an aggregate containing a float16_t, the offset must also be a multiple of 2..."
+        else if (type.containsBasicType(EbtFloat16) && !IsMultipleOfPow2(qualifier.layoutXfbOffset, 2))
+            error(loc, "type contains half float; xfb_offset must be a multiple of 2", "xfb_offset", "");
+#endif
         else if (! IsMultipleOfPow2(qualifier.layoutXfbOffset, 4))
             error(loc, "must be a multiple of size of first component", "xfb_offset", "");
     }
@@ -4662,6 +4667,9 @@ void TParseContext::layoutTypeCheck(const TSourceLoc& loc, const TType& type)
         case EbtBool:
         case EbtFloat:
         case EbtDouble:
+#ifdef AMD_EXTENSIONS
+        case EbtFloat16:
+#endif
             break;
         default:
             error(loc, "cannot be applied to this type", "constant_id", "");
@@ -5561,6 +5569,24 @@ TIntermTyped* TParseContext::constructBuiltIn(const TType& type, TOperator op, T
         basicOp = EOpConstructDouble;
         break;
 
+#ifdef AMD_EXTENSIONS
+    case EOpConstructF16Vec2:
+    case EOpConstructF16Vec3:
+    case EOpConstructF16Vec4:
+    case EOpConstructF16Mat2x2:
+    case EOpConstructF16Mat2x3:
+    case EOpConstructF16Mat2x4:
+    case EOpConstructF16Mat3x2:
+    case EOpConstructF16Mat3x3:
+    case EOpConstructF16Mat3x4:
+    case EOpConstructF16Mat4x2:
+    case EOpConstructF16Mat4x3:
+    case EOpConstructF16Mat4x4:
+    case EOpConstructFloat16:
+        basicOp = EOpConstructFloat16;
+        break;
+#endif
+
     case EOpConstructIVec2:
     case EOpConstructIVec3:
     case EOpConstructIVec4: