void runBeforePass(mlir::Pass* pass, mlir::Operation* op) override {
    // Always print before the first pass.
    if (!printed_) {
      Dump("before_all", op);
      printed_ = true;
    }

    if (RE2::FullMatch(pass->getName(), *dump_pass_re_)) {
      Dump(absl::StrCat(absl::string_view(pass->getName()), "_before"), op,

		Height:   sql.NullFloat64{Float64: 1.8888, Valid: true},
		Birthday: sql.NullTime{Time: time.Now(), Valid: true},
		Allergen: NullString{sql.NullString{String: "Allergen", Valid: true}},
		Password: EncryptedData("pass1"),
		Bytes:    []byte("byte"),
		Num:      18,
		Strings:  StringsSlice{"a", "b", "c"},
		Structs: StructsSlice{
			{"name1", "value1"},
			{"name2", "value2"},
		},
		Role:             Role{Name: "admin"},

                         /*successors=*/false);

  // Locate which variable inputs are part of the forwards pass. These will
  // also be used in the backwards pass. We need to create a 'private' copy
  // of the TpuReplicatedInput for for the fowards pass if there are users
  // outside the pass. Note that in the case of the backwards pass existing
  // this will be the case.
  // This means that when we have put all out sections together some resource

  // Only one of EmbeddingSequencing and EmbeddingPipelining will actually
  // run and the logic is in EmbeddingPipeliningPass. If the pipelining pass
  // runs, embedding attributes are stripped and the sequencing pass will have
  // no effect. If the pipelining pass doesn't run, embedding attributes are
  // preserved and the sequencing rewrite will trigger.
  pm.addPass(mlir::TFDevice::CreateEmbeddingPipeliningPass());

See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
include "mlir/Pass/PassBase.td"

def VerifyClusteringPass : Pass<"verify-clustering-pass", "mlir::func::FuncOp"> {

  let summary = "Verify that the Bridge output is correct and errors if verification fails.";

  let description = [{

}

// Creates an instance of the default quant parameters pass.
std::unique_ptr<OperationPass<func::FuncOp>> CreateImportQuantStatsPass(
    OperationToName op_to_name, const std::string &stats_str) {
  auto pass = std::make_unique<ImportQuantStatsPass>(op_to_name);
  if (pass->ParseQuantStats(stats_str)) return nullptr;
  return pass;
}

// Creates an instance pass to import quantization stats to the operations in

#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/StringRef.h"
#include "mlir/Dialect/Func/IR/FuncOps.h"  // from @llvm-project
#include "mlir/IR/BuiltinOps.h"  // from @llvm-project
#include "mlir/Pass/Pass.h"  // from @llvm-project
#include "mlir/Pass/PassManager.h"  // from @llvm-project
#include "tensorflow/compiler/tf2xla/layout_util.h"
#include "tensorflow/compiler/tf2xla/xla_argument.h"
#include "tensorflow/compiler/tf2xla/xla_helpers.h"

				}
				if f.pass != nil && f.pass.debug > 4 {
					if l == nil {
						fmt.Printf("loop finding    succ %s of %s has no loop\n", bb.String(), b.String())
					} else {
						fmt.Printf("loop finding    succ %s of %s provides loop with header %s\n", bb.String(), b.String(), l.header.String())
					}
				}
			} else { // No loop
				if f.pass != nil && f.pass.debug > 4 {

    return %4 : tensor<2x1xf32>
  }
```

#### Compute Costs Pass
In the compute cost pass, we will essentially compute the cost of each op within
the `FuncOp` based on the target-device cost model and sum them together.


#### Pick Subgraphs Pass
In the pick subgraphs pass, we will pick those subgraphs which can minimize the

		return nil, nil // doesn't use cgo
	}

	cgofiles, info, err := typeCheckCgoSourceFiles(pass.Fset, pass.Pkg, pass.Files, pass.TypesInfo, pass.TypesSizes)
	if err != nil {
		return nil, err
	}
	for _, f := range cgofiles {
		checkCgo(pass.Fset, f, info, pass.Reportf)
	}
	return nil, nil
}

func checkCgo(fset *token.FileSet, f *ast.File, info *types.Info, reportf func(token.Pos, string, ...interface{})) {