wehub-resource-sync
1172 lines
50 KiB
C++
1172 lines
50 KiB
C++
/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License.
|
|
==============================================================================*/
|
|
#include <stdint.h>
|
|
|
|
#include <algorithm>
|
|
#include <array>
|
|
#include <cstddef>
|
|
#include <functional>
|
|
#include <limits>
|
|
#include <numeric>
|
|
#include <random>
|
|
#include <utility>
|
|
#include <vector>
|
|
|
|
#include <gmock/gmock.h>
|
|
#include <gtest/gtest.h>
|
|
#include "flatbuffers/flatbuffers.h" // from @flatbuffers
|
|
#include "tensorflow/lite/kernels/test_util.h"
|
|
#include "tensorflow/lite/schema/schema_generated.h"
|
|
|
|
namespace tflite {
|
|
namespace {
|
|
|
|
using ::testing::ElementsAreArray;
|
|
|
|
class BaseSubOpModel : public SingleOpModel {
|
|
public:
|
|
BaseSubOpModel(const TensorData& input1, const TensorData& input2,
|
|
const TensorData& output,
|
|
ActivationFunctionType activation_type) {
|
|
input1_ = AddInput(input1);
|
|
input2_ = AddInput(input2);
|
|
output_ = AddOutput(output);
|
|
SetBuiltinOp(BuiltinOperator_SUB, BuiltinOptions_SubOptions,
|
|
CreateSubOptions(builder_, activation_type).Union());
|
|
SetBypassDefaultDelegates();
|
|
BuildInterpreter({GetShape(input1_), GetShape(input2_)});
|
|
}
|
|
|
|
int input1() { return input1_; }
|
|
int input2() { return input2_; }
|
|
|
|
void Resize(const std::vector<int>& input1_shape,
|
|
const std::vector<int>& input2_shape) {
|
|
interpreter_->ResizeInputTensor(input1_, input1_shape);
|
|
interpreter_->ResizeInputTensor(input2_, input2_shape);
|
|
AllocateTensors();
|
|
}
|
|
|
|
std::vector<int> GetOutputShape() { return GetTensorShape(output_); }
|
|
|
|
protected:
|
|
int input1_;
|
|
int input2_;
|
|
int output_;
|
|
};
|
|
|
|
template <typename T>
|
|
class SubOpModel : public BaseSubOpModel {
|
|
public:
|
|
using BaseSubOpModel::BaseSubOpModel;
|
|
|
|
std::vector<T> GetOutput() { return ExtractVector<T>(output_); }
|
|
};
|
|
|
|
template <typename T>
|
|
class FloatSubTest : public ::testing::Test {};
|
|
|
|
using FloatSubTestTypes = ::testing::Types<float, half, Eigen::bfloat16>;
|
|
TYPED_TEST_SUITE(FloatSubTest, FloatSubTestTypes);
|
|
|
|
class IntegerSubOpModel : public BaseSubOpModel {
|
|
public:
|
|
using BaseSubOpModel::BaseSubOpModel;
|
|
|
|
template <typename T>
|
|
std::vector<T> GetOutput() {
|
|
return ExtractVector<T>(output_);
|
|
}
|
|
};
|
|
|
|
class QuantizedSubOpModel : public BaseSubOpModel {
|
|
public:
|
|
QuantizedSubOpModel(TensorData input1, TensorData input2, TensorData output,
|
|
ActivationFunctionType activation_type)
|
|
: BaseSubOpModel(SymmetricInt16Scaling(std::move(input1)),
|
|
SymmetricInt16Scaling(std::move(input2)),
|
|
SymmetricInt16Scaling(std::move(output)),
|
|
activation_type) {}
|
|
|
|
template <typename integer_dtype>
|
|
std::vector<float> GetDequantizedOutput() {
|
|
return Dequantize<integer_dtype>(ExtractVector<integer_dtype>(output_),
|
|
GetScale(output_), GetZeroPoint(output_));
|
|
}
|
|
|
|
private:
|
|
TensorData SymmetricInt16Scaling(TensorData tensor) {
|
|
// Symmetric range and null zero-point is required for INT16 tensors. As
|
|
// SingleOpModel::QuantizationParams calculates the scale on an asymmetric
|
|
// base [int_type::min, int_type::max], manually calculate the scale on a
|
|
// symmetric range [int_type::min+1, int_type::max] to ensure a null
|
|
// zero-point.
|
|
if (tensor.type == TensorType_INT16) {
|
|
CHECK_EQ(std::abs(tensor.min), tensor.max);
|
|
tensor.scale = tensor.max / std::numeric_limits<int16_t>::max();
|
|
tensor.zero_point = 0;
|
|
tensor.min = 0;
|
|
tensor.max = 0;
|
|
}
|
|
|
|
return tensor;
|
|
}
|
|
};
|
|
|
|
// for quantized Sub, the error shouldn't exceed step
|
|
template <typename T>
|
|
float GetTolerance(float min, float max) {
|
|
float kQuantizedStep = (max - min) / (std::numeric_limits<T>::max() -
|
|
std::numeric_limits<T>::min());
|
|
return 2.0 * kQuantizedStep;
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, FirstInputZero) {
|
|
using T = TypeParam;
|
|
if (SingleOpModel::GetForceUseNnapi()) {
|
|
return;
|
|
}
|
|
SubOpModel<T> m({GetTensorType<T>(), {0}}, {GetTensorType<T>(), {}},
|
|
{GetTensorType<T>(), {}}, ActivationFunctionType_NONE);
|
|
m.template PopulateTensor<T>(m.input2(), {0.1});
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(m.GetOutputShape(), ElementsAreArray<int>({0}));
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, SecondInputZero) {
|
|
using T = TypeParam;
|
|
if (SingleOpModel::GetForceUseNnapi()) {
|
|
return;
|
|
}
|
|
SubOpModel<T> m({GetTensorType<T>(), {}}, {GetTensorType<T>(), {0}},
|
|
{GetTensorType<T>(), {}}, ActivationFunctionType_NONE);
|
|
m.template PopulateTensor<T>(m.input1(), {0.1});
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(m.GetOutputShape(), ElementsAreArray<int>({0}));
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, NoActivationInplaceInput0) {
|
|
using T = TypeParam;
|
|
SubOpModel<T> m({GetTensorType<T>(), {1, 2, 2, 1}},
|
|
{GetTensorType<T>(), {1, 2, 2, 1}}, {GetTensorType<T>(), {}},
|
|
ActivationFunctionType_NONE);
|
|
const int kInplaceInputTensorIdx = 0;
|
|
const int kInplaceOutputTensorIdx = 0;
|
|
const TfLiteTensor* input_tensor = m.GetInputTensor(kInplaceInputTensorIdx);
|
|
TfLiteTensor* output_tensor = m.GetOutputTensor(kInplaceOutputTensorIdx);
|
|
output_tensor->data.data = input_tensor->data.data;
|
|
m.template PopulateTensor<T>(m.input1(), {-2.0, 0.2, 1.7, 0.5});
|
|
m.template PopulateTensor<T>(m.input2(), {0.1, 0.2, 0.3, 0.8});
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(m.GetOutput(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-2.1, 0.0, 1.4, -0.3},
|
|
static_cast<float>(NumericLimits<T>::epsilon()) * 10)));
|
|
EXPECT_EQ(output_tensor->data.data, input_tensor->data.data);
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, NoActivationInplaceInput1) {
|
|
using T = TypeParam;
|
|
SubOpModel<T> m({GetTensorType<T>(), {1, 2, 2, 1}},
|
|
{GetTensorType<T>(), {1, 2, 2, 1}}, {GetTensorType<T>(), {}},
|
|
ActivationFunctionType_NONE);
|
|
const int kInplaceInputTensorIdx = 1;
|
|
const int kInplaceOutputTensorIdx = 0;
|
|
const TfLiteTensor* input_tensor = m.GetInputTensor(kInplaceInputTensorIdx);
|
|
TfLiteTensor* output_tensor = m.GetOutputTensor(kInplaceOutputTensorIdx);
|
|
output_tensor->data.data = input_tensor->data.data;
|
|
m.template PopulateTensor<T>(m.input1(), {-2.0, 0.2, 1.7, 0.5});
|
|
m.template PopulateTensor<T>(m.input2(), {0.1, 0.2, 0.3, 0.8});
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(m.GetOutput(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-2.1, 0.0, 1.4, -0.3},
|
|
static_cast<float>(NumericLimits<T>::epsilon()) * 10)));
|
|
EXPECT_EQ(output_tensor->data.data, input_tensor->data.data);
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, NoActivation) {
|
|
using T = TypeParam;
|
|
SubOpModel<T> m({GetTensorType<T>(), {1, 2, 2, 1}},
|
|
{GetTensorType<T>(), {1, 2, 2, 1}}, {GetTensorType<T>(), {}},
|
|
ActivationFunctionType_NONE);
|
|
m.template PopulateTensor<T>(m.input1(), {-2.0, 0.2, 1.7, 0.5});
|
|
m.template PopulateTensor<T>(m.input2(), {0.1, 0.2, 0.3, 0.8});
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(m.GetOutput(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-2.1, 0.0, 1.4, -0.3},
|
|
static_cast<float>(NumericLimits<T>::epsilon()) * 10)));
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, ActivationRELU_N1_TO_1) {
|
|
using T = TypeParam;
|
|
SubOpModel<T> m({GetTensorType<T>(), {1, 2, 2, 1}},
|
|
{GetTensorType<T>(), {1, 2, 2, 1}}, {GetTensorType<T>(), {}},
|
|
ActivationFunctionType_RELU_N1_TO_1);
|
|
m.template PopulateTensor<T>(m.input1(), {-2.0, 0.2, 1.7, 0.5});
|
|
m.template PopulateTensor<T>(m.input2(), {0.1, 0.2, 0.3, 0.8});
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(m.GetOutput(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-1.0, 0.0, 1.0, -0.3},
|
|
static_cast<float>(NumericLimits<T>::epsilon()) * 10)));
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, VariousInputShapes) {
|
|
using T = TypeParam;
|
|
const std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
SubOpModel<T> m({GetTensorType<T>(), test_shapes[i]},
|
|
{GetTensorType<T>(), test_shapes[i]},
|
|
{GetTensorType<T>(), {}}, ActivationFunctionType_NONE);
|
|
m.template PopulateTensor<T>(m.input1(), {-2.0, 0.2, 1.7, 0.5, -1.1, 2.0});
|
|
m.template PopulateTensor<T>(m.input2(), {0.1, 0.2, 0.3, 0.8, -1.1, 0.1});
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(m.GetOutput(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-2.1, 0.0, 1.4, -0.3, 0.0, 1.9},
|
|
static_cast<float>(NumericLimits<T>::epsilon()) * 10)))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, WithBroadcast) {
|
|
using T = TypeParam;
|
|
const std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
SubOpModel<T> m({GetTensorType<T>(), test_shapes[i]},
|
|
{GetTensorType<T>(), {}}, // always a scalar
|
|
{GetTensorType<T>(), {}}, ActivationFunctionType_NONE);
|
|
m.template PopulateTensor<T>(m.input1(), {-2.0, 0.2, 1.7, 0.5, -1.1, 2.0});
|
|
m.template PopulateTensor<T>(m.input2(), {0.5});
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(m.GetOutput(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-2.5, -0.3, 1.2, 0.0, -1.6, 1.5},
|
|
static_cast<float>(NumericLimits<T>::epsilon()) * 10)))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, WithBroadcast5D) {
|
|
using T = TypeParam;
|
|
const std::vector<std::vector<int>> test_shapes = {{1, 3, 1, 2, 1}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
SubOpModel<T> m({GetTensorType<T>(), test_shapes[i]},
|
|
{GetTensorType<T>(), {}}, // always a scalar
|
|
{GetTensorType<T>(), {}}, ActivationFunctionType_NONE);
|
|
m.template PopulateTensor<T>(m.input1(), {-2.0, 0.2, 1.7, 0.5, -1.1, 2.0});
|
|
m.template PopulateTensor<T>(m.input2(), {0.5});
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(m.GetOutput(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-2.5, -0.3, 1.2, 0.0, -1.6, 1.5},
|
|
static_cast<float>(NumericLimits<T>::epsilon()) * 10)))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
TYPED_TEST(FloatSubTest, WithBroadcastRankSeven) {
|
|
using T = TypeParam;
|
|
const std::vector<int> input1_shape = {2, 1, 2, 1, 2, 1, 2};
|
|
const std::vector<int> input2_shape = {1, 2, 1, 2, 1, 2, 1};
|
|
const std::vector<int> output_shape = {2, 2, 2, 2, 2, 2, 2};
|
|
SubOpModel<T> m({GetTensorType<T>(), input1_shape},
|
|
{GetTensorType<T>(), input2_shape}, {GetTensorType<T>(), {}},
|
|
ActivationFunctionType_NONE);
|
|
|
|
std::vector<float> input1(16);
|
|
std::vector<float> input2(8);
|
|
std::iota(input1.begin(), input1.end(), 1.0f);
|
|
std::iota(input2.begin(), input2.end(), 0.25f);
|
|
m.template PopulateTensor<T>(m.input1(), ToVector<T>(input1));
|
|
m.template PopulateTensor<T>(m.input2(), ToVector<T>(input2));
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
|
|
auto strides_for = [](const std::vector<int>& shape) {
|
|
std::vector<int> strides(shape.size());
|
|
int stride = 1;
|
|
for (int i = shape.size() - 1; i >= 0; --i) {
|
|
strides[i] = stride;
|
|
stride *= shape[i];
|
|
}
|
|
return strides;
|
|
};
|
|
const std::vector<int> input1_strides = strides_for(input1_shape);
|
|
const std::vector<int> input2_strides = strides_for(input2_shape);
|
|
const std::vector<int> output_strides = strides_for(output_shape);
|
|
std::vector<float> expected(128);
|
|
for (int output_index = 0; output_index < expected.size(); ++output_index) {
|
|
int remaining_index = output_index;
|
|
int input1_index = 0;
|
|
int input2_index = 0;
|
|
for (int dim = 0; dim < output_shape.size(); ++dim) {
|
|
const int coordinate = remaining_index / output_strides[dim];
|
|
remaining_index %= output_strides[dim];
|
|
if (input1_shape[dim] != 1) {
|
|
input1_index += coordinate * input1_strides[dim];
|
|
}
|
|
if (input2_shape[dim] != 1) {
|
|
input2_index += coordinate * input2_strides[dim];
|
|
}
|
|
}
|
|
expected[output_index] = input1[input1_index] - input2[input2_index];
|
|
}
|
|
|
|
EXPECT_THAT(
|
|
m.GetOutput(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
expected, static_cast<float>(NumericLimits<T>::epsilon()) * 10)));
|
|
}
|
|
|
|
TEST(IntegerSubOpModel, NoActivation) {
|
|
IntegerSubOpModel m({TensorType_INT32, {1, 2, 2, 1}},
|
|
{TensorType_INT32, {1, 2, 2, 1}}, {TensorType_INT32, {}},
|
|
ActivationFunctionType_NONE);
|
|
m.PopulateTensor<int32_t>(m.input1(), {-20, 2, 7, 8});
|
|
m.PopulateTensor<int32_t>(m.input2(), {1, 2, 3, 5});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetOutput<int32_t>(), ElementsAreArray({-21, 0, 4, 3}));
|
|
}
|
|
|
|
TEST(IntegerSubOpModel, ActivationRELU_N1_TO_1) {
|
|
IntegerSubOpModel m({TensorType_INT32, {1, 2, 2, 1}},
|
|
{TensorType_INT32, {1, 2, 2, 1}}, {TensorType_INT32, {}},
|
|
ActivationFunctionType_RELU_N1_TO_1);
|
|
m.PopulateTensor<int32_t>(m.input1(), {-20, 2, 7, 8});
|
|
m.PopulateTensor<int32_t>(m.input2(), {1, 2, 3, 5});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetOutput<int32_t>(), ElementsAreArray({-1, 0, 1, 1}));
|
|
}
|
|
|
|
TEST(IntegerSubOpModel, VariousInputShapes) {
|
|
std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
IntegerSubOpModel m({TensorType_INT32, test_shapes[i]},
|
|
{TensorType_INT32, test_shapes[i]},
|
|
{TensorType_INT32, {}}, ActivationFunctionType_NONE);
|
|
m.PopulateTensor<int32_t>(m.input1(), {-20, 2, 7, 8, 11, 20});
|
|
m.PopulateTensor<int32_t>(m.input2(), {1, 2, 3, 5, 11, 1});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetOutput<int32_t>(), ElementsAreArray({-21, 0, 4, 3, 0, 19}))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
TEST(IntegerSubOpModel, WithBroadcast) {
|
|
std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}, {1, 3, 1, 2, 1}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
IntegerSubOpModel m({TensorType_INT32, test_shapes[i]},
|
|
{TensorType_INT32, {}}, // always a scalar
|
|
{TensorType_INT32, {}}, ActivationFunctionType_NONE);
|
|
m.PopulateTensor<int32_t>(m.input1(), {-20, 2, 7, 8, 11, 20});
|
|
m.PopulateTensor<int32_t>(m.input2(), {1});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetOutput<int32_t>(),
|
|
ElementsAreArray(ArrayFloatNear({-21, 1, 6, 7, 10, 19})))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
TEST(Int64SubOpModel, NoActivation) {
|
|
IntegerSubOpModel m({TensorType_INT64, {1, 2, 2, 1}},
|
|
{TensorType_INT64, {1, 2, 2, 1}}, {TensorType_INT64, {}},
|
|
ActivationFunctionType_NONE);
|
|
m.PopulateTensor<int64_t>(m.input1(), {-20, 2, 7, 8});
|
|
m.PopulateTensor<int64_t>(m.input2(), {1, 2, 3, 5});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetOutput<int64_t>(), ElementsAreArray({-21, 0, 4, 3}));
|
|
}
|
|
|
|
TEST(Int64SubOpModel, ActivationRELU_N1_TO_1) {
|
|
IntegerSubOpModel m({TensorType_INT64, {1, 2, 2, 1}},
|
|
{TensorType_INT64, {1, 2, 2, 1}}, {TensorType_INT64, {}},
|
|
ActivationFunctionType_RELU_N1_TO_1);
|
|
m.PopulateTensor<int64_t>(m.input1(), {-20, 2, 7, 8});
|
|
m.PopulateTensor<int64_t>(m.input2(), {1, 2, 3, 5});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetOutput<int64_t>(), ElementsAreArray({-1, 0, 1, 1}));
|
|
}
|
|
|
|
TEST(Int64SubOpModel, VariousInputShapes) {
|
|
std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
IntegerSubOpModel m({TensorType_INT64, test_shapes[i]},
|
|
{TensorType_INT64, test_shapes[i]},
|
|
{TensorType_INT64, {}}, ActivationFunctionType_NONE);
|
|
m.PopulateTensor<int64_t>(m.input1(), {-20, 2, 7, 8, 11, 20});
|
|
m.PopulateTensor<int64_t>(m.input2(), {1, 2, 3, 5, 11, 1});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetOutput<int64_t>(), ElementsAreArray({-21, 0, 4, 3, 0, 19}))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
TEST(Int64SubOpModel, WithBroadcast) {
|
|
std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}, {1, 3, 1, 2, 1}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
IntegerSubOpModel m({TensorType_INT64, test_shapes[i]},
|
|
{TensorType_INT64, {}}, // always a scalar
|
|
{TensorType_INT64, {}}, ActivationFunctionType_NONE);
|
|
m.PopulateTensor<int64_t>(m.input1(), {-20, 2, 7, 8, 11, 20});
|
|
m.PopulateTensor<int64_t>(m.input2(), {1});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetOutput<int64_t>(),
|
|
ElementsAreArray(ArrayFloatNear({-21, 1, 6, 7, 10, 19})))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
template <TensorType tensor_type, typename integer_dtype>
|
|
void QuantizedTestsNoActivation() {
|
|
float kQuantizedTolerance = GetTolerance<integer_dtype>(-0.9, 0.9);
|
|
std::vector<std::vector<float>> inputs1 = {
|
|
{0.1, 0.2, 0.3, 0.4}, {-0.2, 0.2, 0.4, 0.7}, {-0.01, 0.2, 0.7, 0.3}};
|
|
std::vector<std::vector<float>> inputs2 = {
|
|
{0.6, 0.4, 0.3, 0.1}, {0.6, 0.4, 0.5, -0.2}, {0.6, 0.4, -0.18, 0.5}};
|
|
std::vector<std::vector<float>> results = {{-0.5, -0.2, 0.0, 0.3},
|
|
{-0.8, -0.2, -0.1, 0.9},
|
|
{-0.61, -0.2, 0.88, -0.2}};
|
|
for (int i = 0; i < inputs1.size(); ++i) {
|
|
QuantizedSubOpModel m({tensor_type, {1, 2, 2, 1}, -0.7, 0.7},
|
|
{tensor_type, {1, 2, 2, 1}, -0.6, 0.6},
|
|
{tensor_type, {}, -0.9, 0.9},
|
|
ActivationFunctionType_NONE);
|
|
m.QuantizeAndPopulate<integer_dtype>(m.input1(), inputs1[i]);
|
|
m.QuantizeAndPopulate<integer_dtype>(m.input2(), inputs2[i]);
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(
|
|
m.GetDequantizedOutput<integer_dtype>(),
|
|
ElementsAreArray(ArrayFloatNear(results[i], kQuantizedTolerance)))
|
|
<< "With test number " << i;
|
|
}
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedTestsNoActivationUInt8) {
|
|
QuantizedTestsNoActivation<TensorType_UINT8, uint8_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedTestsNoActivationInt8) {
|
|
QuantizedTestsNoActivation<TensorType_INT8, int8_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedTestsNoActivationGenericInt16) {
|
|
QuantizedTestsNoActivation<TensorType_INT16, int16_t>();
|
|
}
|
|
|
|
template <TensorType tensor_type, typename integer_dtype>
|
|
void QuantizedTestsActivationRELU_N1_TO_1() {
|
|
float kQuantizedTolerance = GetTolerance<integer_dtype>(-1.0, 1.0);
|
|
std::vector<std::vector<float>> inputs1 = {{-0.8, 0.2, 0.9, 0.7},
|
|
{-0.8, 0.2, 0.7, 0.5}};
|
|
std::vector<std::vector<float>> inputs2 = {{0.6, 0.4, 0.9, -0.8},
|
|
{0.6, 0.4, -0.8, 0.3}};
|
|
std::vector<std::vector<float>> results = {{-1.0, -0.2, 0.0, 1.0},
|
|
{-1.0, -0.2, 1.0, 0.2}};
|
|
for (int i = 0; i < inputs1.size(); ++i) {
|
|
QuantizedSubOpModel m({tensor_type, {1, 2, 2, 1}, -0.9, 0.9},
|
|
{tensor_type, {1, 2, 2, 1}, -0.9, 0.9},
|
|
{tensor_type, {}, -1.0, 1.0},
|
|
ActivationFunctionType_RELU_N1_TO_1);
|
|
m.QuantizeAndPopulate<integer_dtype>(m.input1(), inputs1[i]);
|
|
m.QuantizeAndPopulate<integer_dtype>(m.input2(), inputs2[i]);
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(
|
|
m.GetDequantizedOutput<integer_dtype>(),
|
|
ElementsAreArray(ArrayFloatNear(results[i], kQuantizedTolerance)))
|
|
<< "With test number " << i;
|
|
}
|
|
}
|
|
TEST(QuantizedSubOpModel, QuantizedTestsActivationRELUN1TO1UInt8) {
|
|
QuantizedTestsActivationRELU_N1_TO_1<TensorType_UINT8, uint8_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedTestsActivationRELUN1TO1Int8) {
|
|
QuantizedTestsActivationRELU_N1_TO_1<TensorType_INT8, int8_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedTestsActivationRELUN1TO1Int16) {
|
|
QuantizedTestsActivationRELU_N1_TO_1<TensorType_INT16, int16_t>();
|
|
}
|
|
|
|
template <TensorType tensor_type, typename integer_dtype>
|
|
void QuantizedVariousInputShapes() {
|
|
float kQuantizedTolerance = GetTolerance<integer_dtype>(-2.1, 2.1);
|
|
std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
QuantizedSubOpModel m({tensor_type, test_shapes[i], -2.0, 2.0},
|
|
{tensor_type, test_shapes[i], -1.1, 1.1},
|
|
{tensor_type, {}, -2.1, 2.1},
|
|
ActivationFunctionType_NONE);
|
|
m.QuantizeAndPopulate<integer_dtype>(m.input1(),
|
|
{-2.0, 0.2, 0.7, 0.8, 1.1, 2.0});
|
|
m.QuantizeAndPopulate<integer_dtype>(m.input2(),
|
|
{0.1, 0.3, 0.3, 0.5, 1.1, 0.1});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetDequantizedOutput<integer_dtype>(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-2.1, -0.1, 0.4, 0.3, 0.0, 1.9}, kQuantizedTolerance)))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedVariousInputShapesUInt8) {
|
|
QuantizedVariousInputShapes<TensorType_UINT8, uint8_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedVariousInputShapesInt8) {
|
|
QuantizedVariousInputShapes<TensorType_INT8, int8_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedVariousInputShapesInt16) {
|
|
QuantizedVariousInputShapes<TensorType_INT16, int16_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedLargeInputShapesInt16) {
|
|
// This test is to cover large shape, which is more than 16 to test
|
|
// AVX2 kernel with batch 16.
|
|
const float kQuantizedTolerance = GetTolerance<int16_t>(-2.1, 2.1);
|
|
const std::vector<int> test_shape = {18};
|
|
QuantizedSubOpModel m({TensorType_INT16, test_shape, -2.0, 2.0},
|
|
{TensorType_INT16, test_shape, -1.1, 1.1},
|
|
{TensorType_INT16, {}, -2.1, 2.1},
|
|
ActivationFunctionType_NONE);
|
|
m.QuantizeAndPopulate<int16_t>(
|
|
m.input1(), {-2.0, 0.2, 0.7, 0.8, 1.1, 2.0, -2.0, 0.2, 0.7, 0.8, 1.1, 2.0,
|
|
-2.0, 0.2, 0.7, 0.8, 1.1, 2.0});
|
|
m.QuantizeAndPopulate<int16_t>(
|
|
m.input2(), {0.1, 0.3, 0.3, 0.5, 1.1, 0.1, 0.1, 0.3, 0.3, 0.5, 1.1, 0.1,
|
|
0.1, 0.3, 0.3, 0.5, 1.1, 0.1});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetDequantizedOutput<int16_t>(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-2.1, -0.1, 0.4, 0.3, 0.0, 1.9, -2.1, -0.1, 0.4, 0.3, 0.0,
|
|
1.9, -2.1, -0.1, 0.4, 0.3, 0.0, 1.9},
|
|
kQuantizedTolerance)));
|
|
}
|
|
|
|
template <TensorType tensor_type, typename integer_dtype>
|
|
void QuantizedWithBroadcast() {
|
|
float kQuantizedTolerance = GetTolerance<integer_dtype>(-2.7, 2.7);
|
|
std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
QuantizedSubOpModel m(
|
|
{tensor_type, test_shapes[i], -2.0, 2.0}, {tensor_type, {}, -0.7, 0.7},
|
|
{tensor_type, {}, -2.7, 2.7}, ActivationFunctionType_NONE);
|
|
m.QuantizeAndPopulate<integer_dtype>(m.input1(),
|
|
{-2.0, 0.2, 0.7, 0.8, 1.1, 2.0});
|
|
m.QuantizeAndPopulate<integer_dtype>(m.input2(), {0.7});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetDequantizedOutput<integer_dtype>(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-2.7, -0.5, 0.0, 0.1, 0.4, 1.3}, kQuantizedTolerance)))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedWithBroadcastUInt8) {
|
|
QuantizedWithBroadcast<TensorType_UINT8, uint8_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedWithBroadcastInt8) {
|
|
QuantizedWithBroadcast<TensorType_INT8, int8_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedWithBroadcastInt16) {
|
|
QuantizedWithBroadcast<TensorType_INT16, int16_t>();
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedTestsNoActivationInt16) {
|
|
float kQuantizedTolerance = GetTolerance<int16_t>(-1.1, 1.1);
|
|
std::vector<std::vector<float>> inputs1 = {
|
|
{0.7, 0.6, 0.6, 0.5}, {-0.2, 0.6, 0.9, -0.1}, {-0.2, 0.6, -0.3, 0.8}};
|
|
std::vector<std::vector<float>> inputs2 = {
|
|
{0.6, 0.4, 0.3, 0.1}, {0.6, 0.4, 0.5, -0.8}, {0.6, 0.4, 0.8, 0.5}};
|
|
std::vector<std::vector<float>> results = {
|
|
{0.1, 0.2, 0.3, 0.4}, {-0.8, 0.2, 0.4, 0.7}, {-0.8, 0.2, -1.1, 0.3}};
|
|
for (int i = 0; i < inputs1.size(); ++i) {
|
|
QuantizedSubOpModel m({TensorType_INT16, {1, 2, 2, 1}, -0.9, 0.9},
|
|
{TensorType_INT16, {1, 2, 2, 1}, -0.8, 0.8},
|
|
{TensorType_INT16, {}, -1.1, 1.1},
|
|
ActivationFunctionType_NONE);
|
|
m.QuantizeAndPopulate<int16_t>(m.input1(), inputs1[i]);
|
|
m.QuantizeAndPopulate<int16_t>(m.input2(), inputs2[i]);
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(
|
|
m.GetDequantizedOutput<int16_t>(),
|
|
ElementsAreArray(ArrayFloatNear(results[i], kQuantizedTolerance)))
|
|
<< "With test number " << i;
|
|
}
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedTestsReluActivationInt16) {
|
|
float kQuantizedTolerance = GetTolerance<int16_t>(-1.0, 1.0);
|
|
std::vector<std::vector<float>> inputs1 = {{-0.8, 0.2, 0.9, 0.7},
|
|
{-0.8, 0.2, 0.7, 0.5}};
|
|
std::vector<std::vector<float>> inputs2 = {{0.6, 0.4, 0.9, -0.8},
|
|
{0.6, 0.4, -0.8, 0.3}};
|
|
std::vector<std::vector<float>> results = {{-1.0, -0.2, 0.0, 1.0},
|
|
{-1.0, -0.2, 1.0, 0.2}};
|
|
for (int i = 0; i < inputs1.size(); ++i) {
|
|
QuantizedSubOpModel m({TensorType_INT16, {1, 2, 2, 1}, -0.9, 0.9},
|
|
{TensorType_INT16, {1, 2, 2, 1}, -0.9, 0.9},
|
|
{TensorType_INT16, {}, -1.0, 1.0},
|
|
ActivationFunctionType_RELU_N1_TO_1);
|
|
m.QuantizeAndPopulate<int16_t>(m.input1(), inputs1[i]);
|
|
m.QuantizeAndPopulate<int16_t>(m.input2(), inputs2[i]);
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(
|
|
m.GetDequantizedOutput<int16_t>(),
|
|
ElementsAreArray(ArrayFloatNear(results[i], kQuantizedTolerance)))
|
|
<< "With test number " << i;
|
|
}
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedTestsNoActivationBroadcastInt16) {
|
|
float kQuantizedTolerance = GetTolerance<int16_t>(-1.1, 1.1);
|
|
std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}, {1, 3, 1, 2, 1}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
QuantizedSubOpModel m({TensorType_INT16, test_shapes[i], -0.9, 0.9},
|
|
{TensorType_INT16, {}, -0.2, 0.2},
|
|
{TensorType_INT16, {}, -1.1, 1.1},
|
|
ActivationFunctionType_NONE);
|
|
m.QuantizeAndPopulate<int16_t>(m.input1(),
|
|
{-0.9, -0.7, -0.3, 0.0, 0.3, 0.5});
|
|
m.QuantizeAndPopulate<int16_t>(m.input2(), {0.2});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetDequantizedOutput<int16_t>(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-1.1, -0.9, -0.5, -0.2, 0.1, 0.3}, kQuantizedTolerance)))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
TEST(QuantizedSubOpModel, QuantizedTestsReluActivationBroadcastInt16) {
|
|
float kQuantizedTolerance = GetTolerance<int16_t>(-1.0, 1.0);
|
|
std::vector<std::vector<int>> test_shapes = {
|
|
{6}, {2, 3}, {2, 1, 3}, {1, 3, 1, 2}, {1, 3, 1, 2, 1}};
|
|
for (int i = 0; i < test_shapes.size(); ++i) {
|
|
QuantizedSubOpModel m({TensorType_INT16, test_shapes[i], -0.9, 0.9},
|
|
{TensorType_INT16, {}, -0.2, 0.2},
|
|
{TensorType_INT16, {}, -1.0, 1.0},
|
|
ActivationFunctionType_RELU_N1_TO_1);
|
|
m.QuantizeAndPopulate<int16_t>(m.input1(),
|
|
{-0.9, -0.7, -0.3, 0.0, 0.3, 0.5});
|
|
m.QuantizeAndPopulate<int16_t>(m.input2(), {0.2});
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetDequantizedOutput<int16_t>(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
{-1.0, -0.9, -0.5, -0.2, 0.1, 0.3}, kQuantizedTolerance)))
|
|
<< "With shape number " << i;
|
|
}
|
|
}
|
|
|
|
constexpr int kDim1 = 2;
|
|
constexpr int kDim2 = 3;
|
|
constexpr int kDim3 = 4;
|
|
constexpr int kDim4 = 5;
|
|
constexpr int kDim5 = 6;
|
|
constexpr int kDim6 = 7;
|
|
|
|
constexpr int kMaxBroadcastDim = 6;
|
|
|
|
template <typename T>
|
|
void TestFloatBroadcast(SubOpModel<T>& m, const std::vector<int>& input1_shape,
|
|
const std::vector<int>& input2_shape) {
|
|
std::array<int, kMaxBroadcastDim> input1_dims;
|
|
std::array<int, kMaxBroadcastDim> input2_dims;
|
|
std::array<int, kMaxBroadcastDim> output_dims;
|
|
std::array<int, kMaxBroadcastDim> input1_strides;
|
|
std::array<int, kMaxBroadcastDim> input2_strides;
|
|
std::array<int, kMaxBroadcastDim> output_strides;
|
|
std::fill(input1_dims.begin(), input1_dims.end(), 1);
|
|
std::fill(input2_dims.begin(), input2_dims.end(), 1);
|
|
std::fill(output_dims.begin(), output_dims.end(), 1);
|
|
std::copy(input1_shape.cbegin(), input1_shape.cend(),
|
|
input1_dims.end() - input1_shape.size());
|
|
std::copy(input2_shape.cbegin(), input2_shape.cend(),
|
|
input2_dims.end() - input2_shape.size());
|
|
|
|
for (size_t i = 0; i < kMaxBroadcastDim; i++) {
|
|
if (input1_dims[i] != 1 && input2_dims[i] != 1) {
|
|
ASSERT_EQ(input1_dims[i], input2_dims[i]);
|
|
}
|
|
output_dims[i] = std::max(input1_dims[i], input2_dims[i]);
|
|
}
|
|
// Compute generalized strides.
|
|
size_t input1_stride = 1, input2_stride = 1, output_stride = 1;
|
|
for (size_t i = kMaxBroadcastDim; i != 0; i--) {
|
|
input1_strides[i - 1] = input1_dims[i - 1] == 1 ? 0 : input1_stride;
|
|
input2_strides[i - 1] = input2_dims[i - 1] == 1 ? 0 : input2_stride;
|
|
output_strides[i - 1] = output_stride;
|
|
input1_stride *= input1_dims[i - 1];
|
|
input2_stride *= input2_dims[i - 1];
|
|
output_stride *= output_dims[i - 1];
|
|
}
|
|
const int num_input1_elements = std::accumulate(
|
|
input1_dims.begin(), input1_dims.end(), 1, std::multiplies<int>());
|
|
const int num_input2_elements = std::accumulate(
|
|
input2_dims.begin(), input2_dims.end(), 1, std::multiplies<int>());
|
|
const int num_output_elements = std::accumulate(
|
|
output_dims.begin(), output_dims.end(), 1, std::multiplies<int>());
|
|
std::vector<T> input1(num_input1_elements);
|
|
std::vector<T> input2(num_input2_elements);
|
|
std::vector<T> output_ref(num_output_elements);
|
|
|
|
std::random_device random_device;
|
|
auto rng = std::mt19937(random_device());
|
|
std::uniform_real_distribution<float> f32dist(0.01f, 1.0f);
|
|
|
|
std::generate(input1.begin(), input1.end(),
|
|
[&]() { return static_cast<T>(f32dist(rng)); });
|
|
std::generate(input2.begin(), input2.end(),
|
|
[&]() { return static_cast<T>(f32dist(rng)); });
|
|
|
|
// Compute reference results.
|
|
for (size_t i = 0; i < output_dims[0]; i++) {
|
|
for (size_t j = 0; j < output_dims[1]; j++) {
|
|
for (size_t k = 0; k < output_dims[2]; k++) {
|
|
for (size_t l = 0; l < output_dims[3]; l++) {
|
|
for (size_t m = 0; m < output_dims[4]; m++) {
|
|
for (size_t n = 0; n < output_dims[5]; n++) {
|
|
output_ref[i * output_strides[0] + j * output_strides[1] +
|
|
k * output_strides[2] + l * output_strides[3] +
|
|
m * output_strides[4] + n * output_strides[5]] =
|
|
static_cast<T>(
|
|
static_cast<float>(
|
|
input1[i * input1_strides[0] + j * input1_strides[1] +
|
|
k * input1_strides[2] + l * input1_strides[3] +
|
|
m * input1_strides[4] +
|
|
n * input1_strides[5]]) -
|
|
static_cast<float>(
|
|
input2[i * input2_strides[0] + j * input2_strides[1] +
|
|
k * input2_strides[2] + l * input2_strides[3] +
|
|
m * input2_strides[4] +
|
|
n * input2_strides[5]]));
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
m.Resize(input1_shape, input2_shape);
|
|
m.template PopulateTensor<T>(m.input1(), input1);
|
|
m.template PopulateTensor<T>(m.input2(), input2);
|
|
TFLITE_INVOKE_AND_CHECK(T, &m);
|
|
EXPECT_THAT(
|
|
m.GetOutput(),
|
|
ElementsAreArray(ArrayFloatNear(
|
|
output_ref, static_cast<float>(NumericLimits<T>::epsilon()) * 10,
|
|
/*fp16_max_abs_err=*/1e-3)));
|
|
}
|
|
|
|
template <typename IntegerType>
|
|
void TestIntegerBroadcast(IntegerSubOpModel& m,
|
|
const std::vector<int>& input1_shape,
|
|
const std::vector<int>& input2_shape) {
|
|
std::array<int, kMaxBroadcastDim> input1_dims;
|
|
std::array<int, kMaxBroadcastDim> input2_dims;
|
|
std::array<int, kMaxBroadcastDim> output_dims;
|
|
std::array<int, kMaxBroadcastDim> input1_strides;
|
|
std::array<int, kMaxBroadcastDim> input2_strides;
|
|
std::array<int, kMaxBroadcastDim> output_strides;
|
|
std::fill(input1_dims.begin(), input1_dims.end(), 1);
|
|
std::fill(input2_dims.begin(), input2_dims.end(), 1);
|
|
std::fill(output_dims.begin(), output_dims.end(), 1);
|
|
std::copy(input1_shape.cbegin(), input1_shape.cend(),
|
|
input1_dims.end() - input1_shape.size());
|
|
std::copy(input2_shape.cbegin(), input2_shape.cend(),
|
|
input2_dims.end() - input2_shape.size());
|
|
|
|
for (size_t i = 0; i < kMaxBroadcastDim; i++) {
|
|
if (input1_dims[i] != 1 && input2_dims[i] != 1) {
|
|
ASSERT_EQ(input1_dims[i], input2_dims[i]);
|
|
}
|
|
output_dims[i] = std::max(input1_dims[i], input2_dims[i]);
|
|
}
|
|
// Compute generalized strides.
|
|
size_t input1_stride = 1, input2_stride = 1, output_stride = 1;
|
|
for (size_t i = kMaxBroadcastDim; i != 0; i--) {
|
|
input1_strides[i - 1] = input1_dims[i - 1] == 1 ? 0 : input1_stride;
|
|
input2_strides[i - 1] = input2_dims[i - 1] == 1 ? 0 : input2_stride;
|
|
output_strides[i - 1] = output_stride;
|
|
input1_stride *= input1_dims[i - 1];
|
|
input2_stride *= input2_dims[i - 1];
|
|
output_stride *= output_dims[i - 1];
|
|
}
|
|
const int num_input1_elements = std::accumulate(
|
|
input1_dims.begin(), input1_dims.end(), 1, std::multiplies<int>());
|
|
const int num_input2_elements = std::accumulate(
|
|
input2_dims.begin(), input2_dims.end(), 1, std::multiplies<int>());
|
|
const int num_output_elements = std::accumulate(
|
|
output_dims.begin(), output_dims.end(), 1, std::multiplies<int>());
|
|
std::vector<IntegerType> input1(num_input1_elements);
|
|
std::vector<IntegerType> input2(num_input2_elements);
|
|
std::vector<IntegerType> output_ref(num_output_elements);
|
|
|
|
std::random_device random_device;
|
|
auto rng = std::mt19937(random_device());
|
|
std::uniform_int_distribution<IntegerType> dist(0, 256);
|
|
|
|
std::generate(input1.begin(), input1.end(), [&]() { return dist(rng); });
|
|
std::generate(input2.begin(), input2.end(), [&]() { return dist(rng); });
|
|
|
|
// Compute reference results.
|
|
for (size_t i = 0; i < output_dims[0]; i++) {
|
|
for (size_t j = 0; j < output_dims[1]; j++) {
|
|
for (size_t k = 0; k < output_dims[2]; k++) {
|
|
for (size_t l = 0; l < output_dims[3]; l++) {
|
|
for (size_t m = 0; m < output_dims[4]; m++) {
|
|
for (size_t n = 0; n < output_dims[5]; n++) {
|
|
output_ref[i * output_strides[0] + j * output_strides[1] +
|
|
k * output_strides[2] + l * output_strides[3] +
|
|
m * output_strides[4] + n * output_strides[5]] =
|
|
input1[i * input1_strides[0] + j * input1_strides[1] +
|
|
k * input1_strides[2] + l * input1_strides[3] +
|
|
m * input1_strides[4] + n * input1_strides[5]] -
|
|
input2[i * input2_strides[0] + j * input2_strides[1] +
|
|
k * input2_strides[2] + l * input2_strides[3] +
|
|
m * input2_strides[4] + n * input2_strides[5]];
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
m.Resize(input1_shape, input2_shape);
|
|
m.PopulateTensor<IntegerType>(m.input1(), input1);
|
|
m.PopulateTensor<IntegerType>(m.input2(), input2);
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
EXPECT_THAT(m.GetOutput<IntegerType>(), testing::ContainerEq(output_ref));
|
|
}
|
|
|
|
// To improve automatic test sharding (via shard_count in the BUILD file),
|
|
// we need to ensure that each individual test case runs in a reasonable time,
|
|
// otherwise we end up being limited by the performance of the longest shard.
|
|
// Since TestFloat32MultiDimBroadcast has 2^12 iterations, it takes a
|
|
// long time (over 30 seconds) to execute all iterations -- too long for a
|
|
// single shard. So we split it into a few \"subshards\" and have a separate
|
|
// TYPED_TEST macro invocation for each subshard.
|
|
|
|
template <typename T>
|
|
void RunFloatMultiDimBroadcastTest(int d1, int d2) {
|
|
const int dims_constants[] = {kDim1, kDim2, kDim3, kDim4, kDim5, kDim6};
|
|
std::vector<int> initial_shape1(d1, 1);
|
|
std::vector<int> initial_shape2(d2, 1);
|
|
SubOpModel<T> m({GetTensorType<T>(), initial_shape1},
|
|
{GetTensorType<T>(), initial_shape2},
|
|
{GetTensorType<T>(), {}}, ActivationFunctionType_NONE);
|
|
|
|
for (uint32_t bm1 = 0; bm1 < (static_cast<uint32_t>(1) << d1); bm1++) {
|
|
for (uint32_t bm2 = 0; bm2 < (static_cast<uint32_t>(1) << d2); bm2++) {
|
|
std::vector<int> input1_shape(d1);
|
|
std::vector<int> input2_shape(d2);
|
|
for (int i = 0; i < d1; ++i) {
|
|
bool broadcast = bm1 & (1 << i);
|
|
input1_shape[i] = broadcast ? 1 : dims_constants[6 - d1 + i];
|
|
}
|
|
for (int i = 0; i < d2; ++i) {
|
|
bool broadcast = bm2 & (1 << i);
|
|
input2_shape[i] = broadcast ? 1 : dims_constants[6 - d2 + i];
|
|
}
|
|
TestFloatBroadcast<T>(m, input1_shape, input2_shape);
|
|
if (testing::Test::IsSkipped()) {
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
#define INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST(d1, d2) \
|
|
TYPED_TEST(FloatSubTest, MultiDimBroadcast_##d1##_##d2) { \
|
|
RunFloatMultiDimBroadcastTest<TypeParam>(d1, d2); \
|
|
}
|
|
|
|
#define INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST_D2(d1) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST(d1, 1) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST(d1, 2) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST(d1, 3) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST(d1, 4) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST(d1, 5) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST(d1, 6)
|
|
|
|
#define INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TESTS() \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST_D2(1) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST_D2(2) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST_D2(3) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST_D2(4) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST_D2(5) \
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TEST_D2(6)
|
|
|
|
INSTANTIATE_FLOAT_SUB_MULTI_DIM_BROADCAST_TESTS()
|
|
|
|
template <typename T>
|
|
class IntegerSubOpTest : public ::testing::Test {};
|
|
|
|
using Int32Or64Types = ::testing::Types<int32_t, int64_t>;
|
|
TYPED_TEST_SUITE(IntegerSubOpTest, Int32Or64Types);
|
|
|
|
// To improve automatic test sharding (via shard_count in the BUILD file),
|
|
// we need to ensure that each individual test case runs in a reasonable time,
|
|
// otherwise we end up being limited by the performance of the longest shard.
|
|
// Since TestIntegerMultiDimBroadcast has 2^12 iterations, it takes a
|
|
// long time (over 30 seconds) to execute all iterations -- too long for a
|
|
// single shard. So we split it into a few "subshards" and have a separate
|
|
// TYPED_TEST macro invocation for each subshard.
|
|
|
|
template <typename TypeParam>
|
|
void RunIntegerMultiDimBroadcastTest(int d1, int d2) {
|
|
const int dims_constants[] = {kDim1, kDim2, kDim3, kDim4, kDim5, kDim6};
|
|
std::vector<int> initial_shape1(d1, 1);
|
|
std::vector<int> initial_shape2(d2, 1);
|
|
IntegerSubOpModel m({GetTensorType<TypeParam>(), initial_shape1},
|
|
{GetTensorType<TypeParam>(), initial_shape2},
|
|
{GetTensorType<TypeParam>(), {}},
|
|
ActivationFunctionType_NONE);
|
|
|
|
for (uint32_t bm1 = 0; bm1 < (static_cast<uint32_t>(1) << d1); bm1++) {
|
|
for (uint32_t bm2 = 0; bm2 < (static_cast<uint32_t>(1) << d2); bm2++) {
|
|
std::vector<int> input1_shape(d1);
|
|
std::vector<int> input2_shape(d2);
|
|
for (int i = 0; i < d1; ++i) {
|
|
bool broadcast = bm1 & (1 << i);
|
|
input1_shape[i] = broadcast ? 1 : dims_constants[6 - d1 + i];
|
|
}
|
|
for (int i = 0; i < d2; ++i) {
|
|
bool broadcast = bm2 & (1 << i);
|
|
input2_shape[i] = broadcast ? 1 : dims_constants[6 - d2 + i];
|
|
}
|
|
TestIntegerBroadcast<TypeParam>(m, input1_shape, input2_shape);
|
|
if (testing::Test::IsSkipped()) {
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
#define INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST(d1, d2) \
|
|
TYPED_TEST(IntegerSubOpTest, MultiDimBroadcast_##d1##_##d2) { \
|
|
RunIntegerMultiDimBroadcastTest<TypeParam>(d1, d2); \
|
|
}
|
|
|
|
#define INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST_D2(d1) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST(d1, 1) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST(d1, 2) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST(d1, 3) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST(d1, 4) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST(d1, 5) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST(d1, 6)
|
|
|
|
#define INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TESTS() \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST_D2(1) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST_D2(2) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST_D2(3) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST_D2(4) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST_D2(5) \
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TEST_D2(6)
|
|
|
|
INSTANTIATE_INTEGER_SUB_MULTI_DIM_BROADCAST_TESTS()
|
|
|
|
template <typename QuantizedType>
|
|
void TestQuantizedBroadcast(QuantizedSubOpModel& m,
|
|
const std::vector<int>& input1_shape,
|
|
const std::vector<int>& input2_shape) {
|
|
std::array<int, kMaxBroadcastDim> input1_dims;
|
|
std::array<int, kMaxBroadcastDim> input2_dims;
|
|
std::array<int, kMaxBroadcastDim> output_dims;
|
|
std::array<int, kMaxBroadcastDim> input1_strides;
|
|
std::array<int, kMaxBroadcastDim> input2_strides;
|
|
std::array<int, kMaxBroadcastDim> output_strides;
|
|
std::fill(input1_dims.begin(), input1_dims.end(), 1);
|
|
std::fill(input2_dims.begin(), input2_dims.end(), 1);
|
|
std::fill(output_dims.begin(), output_dims.end(), 1);
|
|
std::copy(input1_shape.cbegin(), input1_shape.cend(),
|
|
input1_dims.end() - input1_shape.size());
|
|
std::copy(input2_shape.cbegin(), input2_shape.cend(),
|
|
input2_dims.end() - input2_shape.size());
|
|
|
|
for (size_t i = 0; i < kMaxBroadcastDim; i++) {
|
|
if (input1_dims[i] != 1 && input2_dims[i] != 1) {
|
|
ASSERT_EQ(input1_dims[i], input2_dims[i]);
|
|
}
|
|
output_dims[i] = std::max(input1_dims[i], input2_dims[i]);
|
|
}
|
|
// Compute generalized strides.
|
|
size_t input1_stride = 1, input2_stride = 1, output_stride = 1;
|
|
for (size_t i = kMaxBroadcastDim; i != 0; i--) {
|
|
input1_strides[i - 1] = input1_dims[i - 1] == 1 ? 0 : input1_stride;
|
|
input2_strides[i - 1] = input2_dims[i - 1] == 1 ? 0 : input2_stride;
|
|
output_strides[i - 1] = output_stride;
|
|
input1_stride *= input1_dims[i - 1];
|
|
input2_stride *= input2_dims[i - 1];
|
|
output_stride *= output_dims[i - 1];
|
|
}
|
|
const int num_input1_elements = std::accumulate(
|
|
input1_dims.begin(), input1_dims.end(), 1, std::multiplies<int>());
|
|
const int num_input2_elements = std::accumulate(
|
|
input2_dims.begin(), input2_dims.end(), 1, std::multiplies<int>());
|
|
const int num_output_elements = std::accumulate(
|
|
output_dims.begin(), output_dims.end(), 1, std::multiplies<int>());
|
|
std::vector<float> input1(num_input1_elements);
|
|
std::vector<float> input2(num_input2_elements);
|
|
std::vector<float> output_ref(num_output_elements);
|
|
|
|
std::random_device random_device;
|
|
auto rng = std::mt19937(random_device());
|
|
|
|
std::uniform_real_distribution<float> dist(-0.5f, 0.5f);
|
|
|
|
std::generate(input1.begin(), input1.end(), [&]() { return dist(rng); });
|
|
std::generate(input2.begin(), input2.end(), [&]() { return dist(rng); });
|
|
|
|
m.Resize(input1_shape, input2_shape);
|
|
m.QuantizeAndPopulate<QuantizedType>(m.input1(), input1);
|
|
m.QuantizeAndPopulate<QuantizedType>(m.input2(), input2);
|
|
// Compute reference results.
|
|
for (size_t i = 0; i < output_dims[0]; i++) {
|
|
for (size_t j = 0; j < output_dims[1]; j++) {
|
|
for (size_t k = 0; k < output_dims[2]; k++) {
|
|
for (size_t l = 0; l < output_dims[3]; l++) {
|
|
for (size_t m = 0; m < output_dims[4]; m++) {
|
|
for (size_t n = 0; n < output_dims[5]; n++) {
|
|
float x = input1[i * input1_strides[0] + j * input1_strides[1] +
|
|
k * input1_strides[2] + l * input1_strides[3] +
|
|
m * input1_strides[4] + n * input1_strides[5]];
|
|
float y = input2[i * input2_strides[0] + j * input2_strides[1] +
|
|
k * input2_strides[2] + l * input2_strides[3] +
|
|
m * input2_strides[4] + n * input2_strides[5]];
|
|
output_ref[i * output_strides[0] + j * output_strides[1] +
|
|
k * output_strides[2] + l * output_strides[3] +
|
|
m * output_strides[4] + n * output_strides[5]] = x - y;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
for (float& output_value : output_ref) {
|
|
output_value = std::max<float>(output_value, -1.0f);
|
|
output_value = std::min<float>(output_value, 1.0f);
|
|
}
|
|
|
|
ASSERT_EQ(m.Invoke(), kTfLiteOk);
|
|
std::vector<float> output = m.GetDequantizedOutput<QuantizedType>();
|
|
for (size_t i = 0; i < output_dims[0]; i++) {
|
|
for (size_t j = 0; j < output_dims[1]; j++) {
|
|
for (size_t k = 0; k < output_dims[2]; k++) {
|
|
for (size_t l = 0; l < output_dims[3]; l++) {
|
|
for (size_t m = 0; m < output_dims[4]; m++) {
|
|
for (size_t n = 0; n < output_dims[5]; n++) {
|
|
const size_t index =
|
|
i * output_strides[0] + j * output_strides[1] +
|
|
k * output_strides[2] + l * output_strides[3] +
|
|
m * output_strides[4] + n * output_strides[5];
|
|
EXPECT_NEAR(output[index], output_ref[index], 0.6f)
|
|
<< "(i, j, k, l, m, n) = (" << i << ", " << j << ", " << k
|
|
<< ", " << l << ", " << m << ", " << n << ")";
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
template <typename T>
|
|
class QuantizedSubOpTest : public ::testing::Test {};
|
|
|
|
using Int8OrUInt8OrInt16Types = ::testing::Types<int8_t, uint8_t, int16_t>;
|
|
TYPED_TEST_SUITE(QuantizedSubOpTest, Int8OrUInt8OrInt16Types);
|
|
|
|
// To improve automatic test sharding (via shard_count in the BUILD file),
|
|
// we need to ensure that each individual test case runs in a reasonable time,
|
|
// otherwise we end up being limited by the performance of the longest shard.
|
|
// Since TestQuantizedMultiDimBroadcast has 2^12 iterations, it takes a
|
|
// long time (over 30 seconds) to execute all iterations -- too long for a
|
|
// single shard. So we split it into a few "subshards" and have a separate
|
|
// TEST macro invocation for each subshard.
|
|
|
|
template <typename T>
|
|
void RunQuantizedMultiDimBroadcastTest(int d1, int d2) {
|
|
const int dims_constants[] = {kDim1, kDim2, kDim3, kDim4, kDim5, kDim6};
|
|
std::vector<int> initial_shape1(d1, 1);
|
|
std::vector<int> initial_shape2(d2, 1);
|
|
QuantizedSubOpModel m({GetTensorType<T>(), initial_shape1, -0.5f, 0.5f},
|
|
{GetTensorType<T>(), initial_shape2, -0.5f, 0.5f},
|
|
{GetTensorType<T>(), {}, -0.5f, 0.5f},
|
|
ActivationFunctionType_NONE);
|
|
|
|
for (uint32_t bm1 = 0; bm1 < (static_cast<uint32_t>(1) << d1); bm1++) {
|
|
for (uint32_t bm2 = 0; bm2 < (static_cast<uint32_t>(1) << d2); bm2++) {
|
|
std::vector<int> input1_shape(d1);
|
|
std::vector<int> input2_shape(d2);
|
|
for (int i = 0; i < d1; ++i) {
|
|
bool broadcast = bm1 & (1 << i);
|
|
input1_shape[i] = broadcast ? 1 : dims_constants[6 - d1 + i];
|
|
}
|
|
for (int i = 0; i < d2; ++i) {
|
|
bool broadcast = bm2 & (1 << i);
|
|
input2_shape[i] = broadcast ? 1 : dims_constants[6 - d2 + i];
|
|
}
|
|
TestQuantizedBroadcast<T>(m, input1_shape, input2_shape);
|
|
if (testing::Test::IsSkipped()) {
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
#define INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST(T, TypeName, d1, \
|
|
d2) \
|
|
TEST(QuantizedSubOpModel, \
|
|
TypeName##QuantizedMultiDimBroadcast_##d1##_##d2) { \
|
|
RunQuantizedMultiDimBroadcastTest<T>(d1, d2); \
|
|
}
|
|
|
|
#define INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST_D2(T, TypeName, d1) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST(T, TypeName, d1, 1) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST(T, TypeName, d1, 2) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST(T, TypeName, d1, 3) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST(T, TypeName, d1, 4) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST(T, TypeName, d1, 5) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST(T, TypeName, d1, 6)
|
|
|
|
#define INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TESTS(T, TypeName) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST_D2(T, TypeName, 1) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST_D2(T, TypeName, 2) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST_D2(T, TypeName, 3) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST_D2(T, TypeName, 4) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST_D2(T, TypeName, 5) \
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TEST_D2(T, TypeName, 6)
|
|
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TESTS(int8_t, Int8)
|
|
INSTANTIATE_QUANTIZED_SUB_MULTI_DIM_BROADCAST_TESTS(uint8_t, Uint8)
|
|
|
|
} // namespace
|
|
} // namespace tflite
|