custom_op_map_);
  }

  RewritePatternSet patterns(&getContext());
  auto func = getOperation();
  auto* ctx = func.getContext();
  TFL::populateWithGenerated(patterns);
  patterns.add<quant::FoldTrivalRequantizeOp<QuantizeOp>>(ctx);
  patterns.add<PruneUnusedOpsWithSideEffect<TFL::LSTMOp>>(ctx);
  patterns.add<PruneUnusedOpsWithSideEffect<TFL::UnidirectionalSequenceLSTMOp>>(

void PrepareQuantizeDRQPass::runOnOperation() {
  MLIRContext* ctx = &getContext();
  RewritePatternSet patterns(ctx);
  ModuleOp module_op = getOperation();

  populateWithGenerated(patterns);
  patterns.add<PrepareDRQQuantizableOp>(ctx, quant_specs_, op_set_,
                                        enable_per_channel_quantization_);
  FrozenRewritePatternSet frozen_patterns(std::move(patterns));

  }

  const quant::QuantPassSpec quant_params = {
      {quant_specs.verify_numeric, error_tolerance_,
       quant_specs.whole_model_verify, enable_log_if_failed_},
      quant_specs};

  populateWithGenerated(patterns);

  if (quant_specs.weight_quantization || quant_specs.use_fake_quant_num_bits ||
      quant_specs.qdq_conversion_mode ==
          quant::QDQConversionMode::kQDQDynamic) {

  auto func = getOperation();

  // The pattern includes
  // - Converting XlaDotV2Op to EinsumOp
  // - Converting XlaGatherOp to SliceOp
  RewritePatternSet patterns(ctx);
  populateWithGenerated(patterns);

  if (failed(applyPatternsAndFoldGreedily(func, std::move(patterns)))) {
    func.emitError() << "quant-converting-tf-xla-op-to-tf-op failed.";
    signalPassFailure();
  }
}

                      "all can be supported."
                   << "\n";
      return;
    }
    RewritePatternSet patterns(&getContext());
    populateWithGenerated(patterns);
    patterns.add<ConvertTensorListPopBack>(&getContext());
    patterns.add<ConvertTensorListPushBack>(&getContext());
    patterns.add<ConvertVariantAddNOp>(&getContext());

void PreprocessOpPass::runOnOperation() {
  MLIRContext* ctx = &getContext();
  RewritePatternSet patterns(ctx);
  ModuleOp module_op = getOperation();

  populateWithGenerated(patterns);
  patterns.add<PreprocessConstantOp>(ctx, op_set_, quantization_method_,
                                     enable_per_channel_quantization_);
  FrozenRewritePatternSet frozen_patterns(std::move(patterns));

  auto func = getOperation();

  // The pattern includes decomposing batch normalization ops, fusing add/mul
  // with a constant operand to a preceding affine operation.
  RewritePatternSet patterns(ctx);
  populateWithGenerated(patterns);
  patterns.add<RemoveIdentity, ConstantFoldQuantizableOperands>(ctx);
  if (op_set_ != OpSet::XLA) {
    // Convert Einsum into BatchMatMul for non-XLA opsets.

void LiftQuantizableSpotsAsFunctionsPass::runOnOperation() {
  MLIRContext* ctx = &getContext();
  RewritePatternSet patterns(ctx);
  ModuleOp module = getOperation();

  populateWithGenerated(patterns);
  patterns.add<CheckQuantizableOps>(ctx, quant_options_);
  FrozenRewritePatternSet frozen_patterns(std::move(patterns));

  // Iterate over the sorted list of functions to keep the order deterministic.

  if (mode_.empty()) return;

  func::FuncOp func = getOperation();
  MLIRContext *ctx = &getContext();

  // Convert binary ops to BiasAdd ops if possible.
  RewritePatternSet patterns(ctx);
  populateWithGenerated(patterns);
  (void)applyPatternsAndFoldGreedily(func, std::move(patterns));

  // Convert unsupported ops to Flex ops.
  auto tf_dialect = ctx->getLoadedDialect<TF::TensorFlowDialect>();

#include "tensorflow/compiler/mlir/tfr/passes/generated_decompose.inc"

void DecomposeTFOpsPass::ApplyCanonicalization() {
  func::FuncOp func = getOperation();
  RewritePatternSet patterns(&getContext());

  populateWithGenerated(patterns);
  populateCanonicalizationPatterns(func, patterns);

  (void)applyPatternsAndFoldGreedily(func, std::move(patterns));
}

LogicalResult DecomposeTFOpsPass::RewriteUnregisteredTFOps() {