}

        @Override
        public <T> T newInstance(Class<T> implType, Class<? super T> baseClass) {
            // TODO - The baseClass can be inferred from the implType, so attach the serialization constructor onto the GeneratedClass rather than parameterizing and caching here
            try {

 * Read for instance http://docs.codehaus.org/display/MAVEN/Deterministic+Lifecycle+Planning
 * </p>
 * <p>
 * In their current implementation they are "bolted" onto the lifecycle by separating them
 * into TaskSegments. This class represents the execution context of one such task segment.
 * </p>
 * <p>
 * Wise voices have suggested that maybe aggregators shouldn't be bound to the ordinary

    int cY = rhs_shape.size() - 2;
    if (bmm_op.getAdjX()) {
      rY = rhs_shape.size() - 2;
      cY = rhs_shape.size() - 1;
    }

    auto reduce_dim_op = rewriter.create<TFL::ConstOp>(
        bmm_op->getLoc(),
        DenseIntElementsAttr::get(
            RankedTensorType::get({1}, rewriter.getI32Type()), {cY}));
    auto sum_op = rewriter.create<TFL::SumOp>(

        "//tensorflow/python/eager:backprop",
        "//tensorflow/python/framework:test_lib",
        "//tensorflow/python/platform:client_testlib",
    ],
)

gen_op_libraries(
    name = "one_op",
    src = "define_op_template.py",
    deps = [
        "//tensorflow/python/platform:flags",
        "@absl_py//absl:app",
    ],

  // +k8s:conversion-gen=false
  // +optional
  optional string serviceAccount = 9;

  // NodeName is a request to schedule this carp onto a specific node. If it is non-empty,
  // the scheduler simply schedules this carp onto that node, assuming that it fits resource
  // requirements.
  // +optional
  optional string nodeName = 10;

  if (!status.ok()) {
    done(status);
    return;
  }

  // Create a reference to hold onto cpu_tensor until after the literal has
  // been transferred
  TensorReference ref(*cpu_tensor);
  if (UseMultipleStreams()) {
    // Unref the host tensor when the transfer completes.
    // We don't defer the call to done() onto the stream here, and the reasons
    // why this is correct are subtle. We assume that:

	"text/scanner"

	"cmd/asm/internal/flags"
	"cmd/internal/objabi"
	"cmd/internal/src"
)

// Input is the main input: a stack of readers and some macro definitions.
// It also handles #include processing (by pushing onto the input stack)
// and parses and instantiates macro definitions.
type Input struct {
	Stack
	includes        []string
	beginningOfLine bool
	ifdefStack      []bool
	macros          map[string]*Macro

	// +optional
	DeprecatedServiceAccount string `json:"serviceAccount,omitempty" protobuf:"bytes,9,opt,name=serviceAccount"`

	// NodeName is a request to schedule this carp onto a specific node. If it is non-empty,
	// the scheduler simply schedules this carp onto that node, assuming that it fits resource
	// requirements.
	// +optional
	NodeName string `json:"nodeName,omitempty" protobuf:"bytes,10,opt,name=nodeName"`

     * <id>some-snap-repo</id>
     * <url>scp://host/snapshot-path</url>
     * </snapshotRepository>
     * </distributionManagement>
     * ]]></origin>
     * <usage><![CDATA[
     * The repositories onto which artifacts should be deployed.
     * One is for releases, the other for snapshots.
     * ]]></usage>
     * </expression>
     * <expressions>
     *
     * @throws IOException

// ConstOp
//===----------------------------------------------------------------------===//

OpFoldResult ConstOp::fold(FoldAdaptor adaptor) {
  auto operands = adaptor.getOperands();
  (void)operands;
  assert(operands.empty() && "constant has no operands");
  // Return the held attribute value.
  return getValue();
}