Stack []*trace.Frame
	Count uint64
	Time  time.Duration
}

func BuildProfile(prof []ProfileRecord) *profile.Profile {
	p := &profile.Profile{
		PeriodType: &profile.ValueType{Type: "trace", Unit: "count"},
		Period:     1,
		SampleType: []*profile.ValueType{
			{Type: "contentions", Unit: "count"},
			{Type: "delay", Unit: "nanoseconds"},
		},
	}
	locs := make(map[uint64]*profile.Location)

  constexpr int64_t kLookupTableFindCostScale = 8;
  constexpr int64_t kLookupTableFindStringKeyCostScale = 16;

  auto value_type = mlir::cast<mlir::TensorType>(op.getValues().getType());
  auto key_type = mlir::cast<mlir::TensorType>(op.getKeys().getType());

  int64_t output_size = InferTensorSize(context, value_type);

  int64_t cost = kLookupTableFindCostScale * output_size;

        }

        @Nullable
        private PropertyMutator propertyMutatorFor(String propertyName, @Nullable Class<?> valueType) {
            final Pair<String, ? extends Class<?>> key = Pair.of(propertyName, valueType);
            final PropertyMutator cached = mutators.get(key);
            if (cached != null) {
                return cached;
            }

Before using the table you will have to initialize it.  After initialization the
table will be immutable.
  }];

  let arguments = (ins
    I32Attr:$table_id,
    TypeAttr:$key_dtype,
    TypeAttr:$value_dtype
  );

  let results = (outs TFL_ResourceTensor:$out);

  let hasOptions = 1;
}

def TFL_HashtableFindOp: TFL_Op<"hashtable_find", []> {

    val objectType: DataTypeRef

    interface Property : ConfigureAccessor {
        val dataProperty: DataProperty

        override val objectType: DataTypeRef
            get() = dataProperty.valueType
    }

    interface Custom : ConfigureAccessor {
        val customAccessorIdentifier: String
    }

    interface ConfiguringLambdaArgument : ConfigureAccessor

	}

	p, err = profile.Parse(&buf)
	if err != nil {
		t.Fatalf("profile.Parse: %v", err)
	}

	// Expected PeriodType and SampleType.
	periodType := &profile.ValueType{Type: "cpu", Unit: "nanoseconds"}
	sampleType := []*profile.ValueType{
		{Type: "samples", Unit: "count"},
		{Type: "cpu", Unit: "nanoseconds"},
	}

	checkProfile(t, p, 2000*1000, periodType, sampleType, nil, "")
}

func f1() { f1() }

    auto values_type = RankedTensorType::get(
        output_shape,
        mlir::cast<quant::UniformQuantizedType>(output_type.getElementType())
            .getStorageType());
    rewriter.replaceOpWithNewOp<QConstOp>(
        op, TypeAttr::get(result_type),
        DenseIntElementsAttr::get(values_type, new_values));
    return success();
  }
};

internal::ParamGenerator<
  typename ::testing::internal::IteratorTraits<ForwardIterator>::value_type>
ValuesIn(ForwardIterator begin, ForwardIterator end);

template <typename T, size_t N>
internal::ParamGenerator<T> ValuesIn(const T (&array)[N]);

template <class Container>
internal::ParamGenerator<typename Container::value_type> ValuesIn(
    const Container& container);

namespace internal {

	keepFramesX        int64
	stringTable        []string
	defaultSampleTypeX int64
}

// ValueType corresponds to Profile.ValueType
type ValueType struct {
	Type string // cpu, wall, inuse_space, etc
	Unit string // seconds, nanoseconds, bytes, etc

	typeX int64
	unitX int64
}

// Sample corresponds to Profile.Sample
type Sample struct {
	Location []*Location
	Value    []int64
	Label    map[string][]string

    <T> DefaultListProperty<T> listProperty(Class<T> elementType);

    <T> DefaultSetProperty<T> setProperty(Class<T> elementType);

    <V, K> DefaultMapProperty<K, V> mapProperty(Class<K> keyType, Class<V> valueType);