absl::Span<const int64_t> shape, const TF_DataType dtype)
      : device_(device),
        tensors_(std::move(tensors)),
        shape_(std::vector<int64_t>(shape.begin(), shape.end())),
        dtype_(dtype) {}
  ParallelTensor(const ParallelDevice& device,
                 std::vector<TensorHandlePtr> tensors, const TF_DataType dtype)
      : device_(device),
        tensors_(std::move(tensors)),

}

func panicmakeslicecap() {
	panic(errorString("makeslice: cap out of range"))
}

// makeslicecopy allocates a slice of "tolen" elements of type "et",
// then copies "fromlen" elements of type "et" into that new allocation from "from".
func makeslicecopy(et *_type, tolen int, fromlen int, from unsafe.Pointer) unsafe.Pointer {
	var tomem, copymem uintptr
	if uintptr(tolen) > uintptr(fromlen) {
		var overflow bool

      // Return values
      {"y: float"},
      // Attr def
      {},
      // Nodes
      {
          {{"two"},
           "Const",
           {},
           {{"value", test::AsScalar<int32>(2)}, {"dtype", DT_INT64}}},
          {{"scale"},
           "Cast",
           {"two"},
           {{"SrcT", DT_INT64}, {"DstT", DT_FLOAT}}},
          {{"y"}, "Mul", {"x", "scale"}, {{"T", DT_FLOAT}}},
      });

		config = &defaultConfig
	}

	x, xType, ok := functionAndType(f)
	if !ok {
		return SetupError("f is not a function")
	}
	y, yType, ok := functionAndType(g)
	if !ok {
		return SetupError("g is not a function")
	}

	if xType != yType {
		return SetupError("functions have different types")
	}

	arguments := make([]reflect.Value, xType.NumIn())
	rand := config.getRand()

      const Tensor* value = variable.var()->tensor();
      arg.type = value->dtype();
      arg.shape = value->shape();
      arg.initialized = true;
    } else {
      // Instead of embedding constant into HLO,
      // we handle tf.constant as parameter to reduce size.
      arg.kind = XlaCompiler::Argument::kParameter;
      arg.type = input->dtype();
      arg.shape = input->shape();
    }
    args.push_back(arg);

  }

  tensorflow::DataType dtype;
  auto status = tensorflow::ConvertToDataType(operand.getType(), &dtype);
  if (!status.ok()) {
    op->emitRemark() << "skipping legalization due to " << status.ToString();
    return tensorflow::XlaExpression::Invalid();
  }
  return tensorflow::XlaExpression::XlaOp(xla_op, dtype);
}

}  // namespace mhlo

	// a pointer/scalar bitmap. We also reserve space for a dummy _type that
	// refers to the bitmap. The PtrBytes field of the dummy _type indicates how
	// many of those bits are valid.
	return userArenaChunkBytes/goarch.PtrSize/8 + unsafe.Sizeof(_type{})
}

type userArena struct {
	// full is a list of full chunks that have not enough free memory left, and

		aType := string(a.Type)
		if aType == "" {
			allErrs = append(allErrs, field.Required(fldPath.Child("type"), ""))
			continue
		}
		if !knownTypes.Has(aType) {
			allErrs = append(allErrs, field.NotSupported(fldPath.Child("type"), aType, knownTypes.List()))
			continue
		}
		if seenAuthorizerTypes.Has(aType) && !repeatableTypes.Has(aType) {

  ctx.appendDialectRegistry(registry);
  ctx.loadAllAvailableDialects();
}

Status ConvertDataTypeToTensor(tensorflow::DataType dtype, Builder builder,
                               Type* type) {
  Status s = tensorflow::ConvertDataType(dtype, builder, type);
  if (s.ok()) *type = UnrankedTensorType::get(*type);
  return s;
}

class MlirTensor : public TracingTensorHandle {
 public:
  explicit MlirTensor(Value value)

	if list == nil {
		return
	}

	var named, anonymous bool

	var typ Type
	var prev syntax.Expr
	for i, field := range list {
		ftype := field.Type
		// type-check type of grouped fields only once
		if ftype != prev {
			prev = ftype
			if t, _ := ftype.(*syntax.DotsType); t != nil {
				ftype = t.Elem
				if variadicOk && i == len(list)-1 {
					variadic = true
				} else {