// If the user is not in one of the regions, we are not interested in it.
      // Since all the sub-regions within this region (i.e., regions attached to
      // op's in this region) have themselves gone through lifting, all resource
      // users are expected to be operations in this region and not embedded
      // within other sub-regions attached to ops in this region. So the check

func (s *gState) endRegion(r userRegion) error {
	if len(s.regions) == 0 {
		// We do not know about regions that began before tracing started.
		return nil
	}
	if next := s.regions[len(s.regions)-1]; next != r {
		return fmt.Errorf("misuse of region in goroutine %v: region end %v when the inner-most active region start event is %v", s.id, r, next)
	}
	s.regions = s.regions[:len(s.regions)-1]
	return nil
}

     }
  ```

  `cond` is the condition region and `body` is the body region. Both these
  regions accept the current value of the iteration variables as inputs.

  The condition region yields a tensor<i1> which, if false, will exit the loop.
  It can also, optionally and additionally, yield the iteration variables, which
  must be unchanged.

  The body region always has to yield the (possibly updated) iteration variables.

  return subgraph.name;
}

// Adds a CallOp in `region` to call the `func` and returns the results of
// CallOp.
void AddCallOpInWhileOpRegion(mlir::Region& region, mlir::func::FuncOp func) {
  OpBuilder op_builder{region};
  region.push_back(new mlir::Block());
  Location loc = region.getLoc();
  auto inputs = func.getFunctionType().getInputs();
  region.addArguments(inputs, mlir::SmallVector<Location>(inputs.size(), loc));

    }
  }
}

// Since we have the outputs from host and device computation after moving
// outside compiled ops, we can create the actual parallel_execute regions.
// Still, one region is for the host computation for outside compilation and
// the other one is for the original Device cluster computation.
mlir::tf_device::ParallelExecuteOp CreateFinalParallelExecuteOp(

    }
  }

  // Ensure that all ops are in the same region, and have the same replication
  // info.
  // TODO(bfontain): Allow for multiple regions/loops in one module.
  // TODO(patn): move this pass after cluster formation to remove the
  // complexity with replication info and metadata, cluster checking and
  // generalizing to multiple TPU clusters.
  Region* region = (*forward_pass_ops.begin())->getParentRegion();

		// against C code, it's possible global constructors
		// have called malloc and adjusted the process' brk.
		// Query the brk so we can avoid trying to map the
		// region over it (which will cause the kernel to put
		// the region somewhere else, likely at a high
		// address).
		procBrk := sbrk0()

		// If we ask for the end of the data segment but the
		// operating system requires a little more space

  "tf.NoOp"() {_xla_compile_device_type = "TPU", _replication_info = "cluster_test_fn"} : () -> ()
  func.return
}


// Test cluster formation with ops with attached regions within a cluster.
// Nested op's that are moved should get their _replication_info and device
// attributes cleared.
// CHECK-LABEL: func @cluster_ops_with_regions
func.func @cluster_ops_with_regions() {

    return success();
  };
};

#include "tensorflow/compiler/mlir/lite/transforms/generated_prepare_tf.inc"

// Returns success if all the operations in the `op`'s regions including `op`
// itself are legal in a TFLite pipeline.
LogicalResult ValidateOp(Operation *op) {
  bool has_illegal_ops = false;
  op->walk([&](Operation *op) {
    if (isa<TF::VariableV2Op>(op)) {

    // Inline the regions from the old while into the new one, and apply
    // signature conversion to inlined region.
    for (auto it : llvm::zip(op.getRegions(), converted.getRegions())) {
      Region &old_region = *std::get<0>(it);
      Region &new_region = *std::get<1>(it);

      Block &entry = old_region.front();
      // Build signature conversion for the region.