RequirementMatcher matcher = provides.get(key);

            if (matcher == null) {
                getLog().debug("Toolchain {} is missing required property: {}", this, key);
                return false;
            }
            if (!matcher.matches(requirement.getValue())) {
                getLog().debug("Toolchain {} doesn't match required property: {}", this, key);
                return false;
            }
        }

            const_op->getLoc(),
            DenseElementsAttr::get(
                tensorflow::GetTypeFromTFTensorShape(
                    {}, splat_elements_attr.getType().getElementType()),
                splat_elements_attr.getSplatValue<Attribute>()));
    TFL::FillOp fill = op_builder->create<TFL::FillOp>(
        const_op->getLoc(), splat_elements_attr.getType(), fill_shape,
        fill_value);

  auto launch_op = builder.create<tf_device::LaunchOp>(
      op->getLoc(), builder.getStringAttr(device),
      /*result_types=*/op->getResultTypes());
  op->replaceAllUsesWith(launch_op);

  launch_op.getBody().push_back(new Block);
  builder.setInsertionPointToEnd(&launch_op.GetBody());
  auto* return_op =
      builder.create<tf_device::ReturnOp>(op->getLoc(), op->getResults())
          .getOperation();

    auto num_parallel_calls_op = rewriter.create<CastOp>(
        op.getLoc(), UnrankedTensorType::get(rewriter.getIntegerType(64)),
        batchInputOp.getNumParallelCalls(), rewriter.getBoolAttr(false));

    if (op.getMetadata() != batchInputOp.getMetadata()) {
      return failure();
    }

    auto fused_op = rewriter.create<MapAndBatchDatasetOp>(
        op.getLoc(), op.getType(), batchInputOp.getInputDataset(),

    } else {
      initial_value = mlir::Attribute();
    }
    opToName[globalTensor] = name;
    auto variableOp = globalBuilder.create<ml_program::GlobalOp>(
        globalTensor.getLoc(), name, globalTensor.getType(),
        globalTensor.getIsMutable(), initial_value,
        /*visibility=*/globalBuilder.getStringAttr("private"));
    variableOp.setPrivate();
  }

  SymbolTable syms(module);

  SendTPUEmbeddingGradientsOp send_op;
  if (failed(GetOp(region, &send_op))) return failure();

  // No TPU embedding ops.
  if (!recv_op && !send_op) return success();

  Location loc = recv_op ? recv_op.getLoc() : send_op.getLoc();
  StringRef config = recv_op ? recv_op.getConfig() : send_op.getConfig();

  // Create XlaRecvTPUEmbeddingDeduplicationData op.
  OpBuilder builder(region);
  auto output_ty =

  void runOnOperation() override;
};

void ReplaceClusterReturnWithReturn(tf_device::ReturnOp cluster_return_op,
                                    OpBuilder* builder) {
  builder->create<func::ReturnOp>(cluster_return_op.getLoc(),
                                  cluster_return_op.getOperands());
  cluster_return_op.erase();
}

// Builds a function that outlines region attached to cluster_op or launch_op,

     * discovery of new URLs on very large sites.
     */
    private String lastmod;

    /*
     * (non-Javadoc)
     *
     * @see org.codelibs.fess.crawler.entity.Sitemap#getLoc()
     */
    @Override
    public String getLoc() {
        return loc;
    }

    public void setLoc(final String loc) {
        this.loc = loc;
    }

    /*
     * (non-Javadoc)
     *

  // Every ModuleOp has at least one region and one block.
  Block* first_block = &module->getRegion(0).front();
  builder.setInsertionPointToEnd(first_block);
  auto func = builder.create<mlir::func::FuncOp>(
      ops[0]->getLoc(), dialect.str() + std::to_string(function_id),
      builder.getFunctionType(input_types, output_types));
  func->setAttr("dialect", builder.getStringAttr(dialect));
  auto block = func.addEntryBlock();

    // TODO: Map to a qbarrier with an attribute like [Forced] to signal that
    // this is a forced/hard-coded constraint.
    auto qbarrier = rewriter.create<QuantizeCastOp>(op.getLoc(), quantizedType,
                                                    op.getInputs());
    rewriter.replaceOpWithNewOp<DequantizeCastOp>(op, converter.input_type,