}
        break;
      }
      case ::tensorflow::AttrValue::kI:
        flex_builder->Int(key, attr.i());
        break;
      case ::tensorflow::AttrValue::kF:
        flex_builder->Float(key, attr.f());
        break;
      case ::tensorflow::AttrValue::kB:
        flex_builder->Bool(key, attr.b());
        break;
      case tensorflow::AttrValue::kList:

    Operation* filter_op = filter_value.getDefiningOp();
    auto filter_uniform_quantized_type =
        GetElementType(filter_value).cast<UniformQuantizedPerAxisType>();
    auto filter_constant_value_attr = cast<DenseIntElementsAttr>(
        cast<stablehlo::ConstantOp>(filter_value.getDefiningOp()).getValue());
    const DenseIntElementsAttr new_filter_value_attr =

    if (!mlir::cast<ShapedType>(conv_op.getRhs().getType()).hasStaticShape()) {
      return false;
    }
    if (!mlir::cast<ShapedType>(conv_op.getLhs().getType()).hasStaticShape() &&
        !mlir::cast<ShapedType>(conv_op.getType()).hasStaticShape()) {
      auto dnums = conv_op.getDimensionNumbers();
      auto lhs_type = mlir::cast<ShapedType>(conv_op.getLhs().getType());

      // to match the input dimensions. In case both the dimensions are one,
      // reducing the dimension has no effect. We choose to reduce such
      // dimensions to match the TensorFlow kernel behavior. However, note that
      // the TF behavior in this case is inconsistent with the case with the
      // same shapes.
      r0.push_back(idx);
      r1.push_back(idx);
    }

			var wantParens bool
			switch {
			case op == nil:
				wantParens = true
			case op.Name == `operator"" `:
				wantParens = false
			case op.Name == "&":
				wantParens = false
			case isDelete:
				wantParens = false
			case op.Name == "alignof ":
				wantParens = true
			case op.Name == "sizeof ":
				wantParens = true
			case op.Name == "typeid ":
				wantParens = true

  }

  // Get the tensor types.
  const auto input_type = input_tensor.getType().cast<ShapedType>();
  const auto weights_type = weights_tensor.getType().cast<ShapedType>();
  const auto bias_type =
      has_bias ? bias_tensor.getType().cast<ShapedType>() : ShapedType{};

  const auto output_type = getType(0).cast<ShapedType>();

  // Folding only implemented for float tensors.

		}

	case C_PSAUTO_4:
		switch b {
		case C_ZAUTO, C_PSAUTO_16, C_PSAUTO_8:
			return true
		}

	case C_PSAUTO:
		switch b {
		case C_ZAUTO, C_PSAUTO_16, C_PSAUTO_8, C_PSAUTO_4:
			return true
		}

	case C_PPAUTO_16:
		switch b {
		case C_ZAUTO, C_PSAUTO_16:
			return true
		}

	case C_PPAUTO:
		switch b {
		case C_ZAUTO, C_PSAUTO_16, C_PSAUTO_8, C_PPAUTO_16:

	case C_S16CON:
		return cmp(C_U15CON, b)
	case C_U16CON:
		return cmp(C_U15CON, b)
	case C_16CON:
		return cmp(C_S16CON, b) || cmp(C_U16CON, b)
	case C_U31CON:
		return cmp(C_U16CON, b)
	case C_U32CON:
		return cmp(C_U31CON, b)
	case C_S32CON:
		return cmp(C_U31CON, b) || cmp(C_S16CON, b)
	case C_32CON:
		return cmp(C_S32CON, b) || cmp(C_U32CON, b)
	case C_S34CON:

func usesLibcall() bool {
	switch GOOS {
	case "aix", "darwin", "illumos", "ios", "solaris", "windows":
		return true
	case "openbsd":
		return GOARCH != "mips64"
	}
	return false
}

// mStackIsSystemAllocated indicates whether this runtime starts on a
// system-allocated stack.
func mStackIsSystemAllocated() bool {
	switch GOOS {

	v.mustBeAssignable()
	switch v.kind() {
	case Bool:
		*(*bool)(v.ptr) = false
	case Int:
		*(*int)(v.ptr) = 0
	case Int8:
		*(*int8)(v.ptr) = 0
	case Int16:
		*(*int16)(v.ptr) = 0
	case Int32:
		*(*int32)(v.ptr) = 0
	case Int64:
		*(*int64)(v.ptr) = 0
	case Uint:
		*(*uint)(v.ptr) = 0
	case Uint8:
		*(*uint8)(v.ptr) = 0
	case Uint16:
		*(*uint16)(v.ptr) = 0
	case Uint32: