STT-tensorflow/tensorflow/lite/delegates/xnnpack/conv_2d_tester.cc
Marat Dukhan a11003dcc3 Support sparse FP16 weights in XNNPACK delegate
PiperOrigin-RevId: 344897498
Change-Id: Ia474dc338321ccfdbdd68d624fa05dc0e9bb9ba5
2020-11-30 15:14:40 -08:00

347 lines
15 KiB
C++

/* Copyright 2020 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 "tensorflow/lite/delegates/xnnpack/conv_2d_tester.h"
#include <array>
#include <cstdint>
#include <functional>
#include <random>
#include <vector>
#include <gtest/gtest.h>
#include <fp16.h>
#include "flatbuffers/flatbuffers.h" // from @flatbuffers
#include "tensorflow/lite/interpreter.h"
#include "tensorflow/lite/kernels/register.h"
#include "tensorflow/lite/model.h"
#include "tensorflow/lite/schema/schema_conversion_utils.h"
#include "tensorflow/lite/schema/schema_generated.h"
#include "tensorflow/lite/version.h"
namespace tflite {
namespace xnnpack {
void Conv2DTester::Test(TfLiteDelegate* delegate) const {
std::vector<char> buffer = CreateTfLiteModel();
const Model* model = GetModel(buffer.data());
std::unique_ptr<Interpreter> delegate_interpreter;
ASSERT_EQ(
InterpreterBuilder(
model,
::tflite::ops::builtin::BuiltinOpResolverWithoutDefaultDelegates())(
&delegate_interpreter),
kTfLiteOk);
std::unique_ptr<Interpreter> default_interpreter;
ASSERT_EQ(
InterpreterBuilder(
model,
::tflite::ops::builtin::BuiltinOpResolverWithoutDefaultDelegates())(
&default_interpreter),
kTfLiteOk);
ASSERT_TRUE(delegate_interpreter);
ASSERT_TRUE(default_interpreter);
ASSERT_EQ(delegate_interpreter->inputs().size(), 1);
ASSERT_EQ(default_interpreter->inputs().size(), 1);
ASSERT_EQ(delegate_interpreter->outputs().size(), 1);
ASSERT_EQ(default_interpreter->outputs().size(), 1);
ASSERT_EQ(delegate_interpreter->AllocateTensors(), kTfLiteOk);
ASSERT_EQ(default_interpreter->AllocateTensors(), kTfLiteOk);
ASSERT_EQ(delegate_interpreter->ModifyGraphWithDelegate(delegate), kTfLiteOk);
std::random_device random_device;
auto rng = std::mt19937(random_device());
auto input_rng =
std::bind(std::uniform_real_distribution<float>(), std::ref(rng));
float* default_input_data = default_interpreter->typed_tensor<float>(
default_interpreter->inputs()[0]);
std::generate(default_input_data,
default_input_data + BatchSize() * InputHeight() *
InputWidth() * InputChannels(),
input_rng);
float* delegate_input_data = delegate_interpreter->typed_tensor<float>(
delegate_interpreter->inputs()[0]);
std::copy(default_input_data,
default_input_data +
BatchSize() * InputHeight() * InputWidth() * InputChannels(),
delegate_input_data);
ASSERT_EQ(default_interpreter->Invoke(), kTfLiteOk);
ASSERT_EQ(delegate_interpreter->Invoke(), kTfLiteOk);
float* default_output_data = default_interpreter->typed_tensor<float>(
default_interpreter->outputs()[0]);
float* delegate_output_data = delegate_interpreter->typed_tensor<float>(
delegate_interpreter->outputs()[0]);
for (int32_t i = 0; i < BatchSize(); i++) {
for (int32_t y = 0; y < OutputHeight(); y++) {
for (int32_t x = 0; x < OutputWidth(); x++) {
for (int32_t c = 0; c < OutputChannels(); c++) {
const int32_t index = ((i * OutputHeight() + y) * OutputWidth() + x) *
OutputChannels() +
c;
ASSERT_NEAR(default_output_data[index], delegate_output_data[index],
std::abs(default_output_data[index]) * 3.0e-6f)
<< "batch " << i << " / " << BatchSize() << ", y position " << y
<< " / " << OutputHeight() << ", x position " << x << " / "
<< OutputWidth() << ", channel " << c << " / "
<< OutputChannels();
}
}
}
}
}
std::vector<char> Conv2DTester::CreateTfLiteModel() const {
std::random_device random_device;
auto rng = std::mt19937(random_device());
auto range_rng = std::bind(
std::uniform_real_distribution<float>(-25.0f, 25.0f), std::ref(rng));
flatbuffers::FlatBufferBuilder builder;
std::vector<flatbuffers::Offset<OperatorCode>> operator_codes{
{CreateOperatorCode(builder, BuiltinOperator_CONV_2D)}};
std::vector<flatbuffers::Offset<tflite::Operator>> operators;
std::vector<flatbuffers::Offset<tflite::Buffer>> buffers{
{CreateBuffer(builder, builder.CreateVector({}))}};
if (SparseWeights()) {
operator_codes.emplace_back(
CreateOperatorCode(builder, BuiltinOperator_DENSIFY));
const std::array<int32_t, 1> densify_filter_inputs{{0}};
const std::array<int32_t, 1> densify_filter_outputs{
{FP16Weights() ? 1 : 2}};
operators.emplace_back(CreateOperator(
builder, /*opcode_index=*/operator_codes.size() - 1,
builder.CreateVector<int32_t>(densify_filter_inputs.data(),
densify_filter_inputs.size()),
builder.CreateVector<int32_t>(densify_filter_outputs.data(),
densify_filter_outputs.size())));
}
if (FP16Weights()) {
operator_codes.emplace_back(
CreateOperatorCode(builder, BuiltinOperator_DEQUANTIZE));
std::vector<uint16_t> filter_data(OutputChannels() * KernelHeight() *
KernelWidth() * InputChannels());
std::vector<uint16_t> bias_data(OutputChannels());
for (int32_t oc = 0; oc < OutputChannels(); oc++) {
// Use the same range of all-positive or all-negative values to generate
// all weights within the same output channel, but different ranges for
// different output channels. This ensures that no catastrophic
// cancellation occur, but test covers both positive and negative inputs.
const float range = range_rng();
auto value_rng =
std::bind(fp16_ieee_from_fp32_value,
std::bind(std::uniform_real_distribution<float>(
std::min(range, 0.0f), std::max(range, 0.0f)),
std::ref(rng)));
bias_data[oc] = value_rng();
for (int32_t ic = 0; ic < InputChannels(); ic++) {
for (int32_t y = 0; y < KernelHeight(); y++) {
for (int32_t x = 0; x < KernelWidth(); x++) {
const int32_t index =
((oc * KernelHeight() + y) * KernelWidth() + x) *
InputChannels() +
ic;
filter_data[index] = value_rng();
}
}
}
}
buffers.emplace_back(CreateBuffer(
builder, builder.CreateVector(
reinterpret_cast<const uint8_t*>(filter_data.data()),
sizeof(uint16_t) * filter_data.size())));
buffers.emplace_back(CreateBuffer(
builder,
builder.CreateVector(reinterpret_cast<const uint8_t*>(bias_data.data()),
sizeof(uint16_t) * bias_data.size())));
const std::array<int32_t, 1> dequantize_filter_inputs{
{SparseWeights() ? 1 : 0}};
const std::array<int32_t, 1> dequantize_filter_outputs{
{SparseWeights() ? 4 : 3}};
operators.emplace_back(CreateOperator(
builder, /*opcode_index=*/operator_codes.size() - 1,
builder.CreateVector<int32_t>(dequantize_filter_inputs.data(),
dequantize_filter_inputs.size()),
builder.CreateVector<int32_t>(dequantize_filter_outputs.data(),
dequantize_filter_outputs.size())));
const std::array<int32_t, 1> dequantize_bias_inputs{
{SparseWeights() ? 2 : 1}};
const std::array<int32_t, 1> dequantize_bias_outputs{
{SparseWeights() ? 5 : 4}};
operators.emplace_back(CreateOperator(
builder, /*opcode_index=*/operator_codes.size() - 1,
builder.CreateVector<int32_t>(dequantize_bias_inputs.data(),
dequantize_bias_inputs.size()),
builder.CreateVector<int32_t>(dequantize_bias_outputs.data(),
dequantize_bias_outputs.size())));
} else {
std::vector<float> filter_data(OutputChannels() * KernelHeight() *
KernelWidth() * InputChannels());
std::vector<float> bias_data(OutputChannels());
for (int32_t oc = 0; oc < OutputChannels(); oc++) {
// Use the same range of all-positive or all-negative values to generate
// all weights within the same output channel, but different ranges for
// different output channels. This ensures that no catastrophic
// cancellation occur, but test covers both positive and negative inputs.
const float range = range_rng();
auto value_rng =
std::bind(std::uniform_real_distribution<float>(
std::min(range, 0.0f), std::max(range, 0.0f)),
std::ref(rng));
bias_data[oc] = value_rng();
for (int32_t ic = 0; ic < InputChannels(); ic++) {
for (int32_t y = 0; y < KernelHeight(); y++) {
for (int32_t x = 0; x < KernelWidth(); x++) {
const int32_t index =
((oc * KernelHeight() + y) * KernelWidth() + x) *
InputChannels() +
ic;
filter_data[index] = value_rng();
}
}
}
}
buffers.emplace_back(CreateBuffer(
builder, builder.CreateVector(
reinterpret_cast<const uint8_t*>(filter_data.data()),
sizeof(float) * filter_data.size())));
buffers.emplace_back(CreateBuffer(
builder,
builder.CreateVector(reinterpret_cast<const uint8_t*>(bias_data.data()),
sizeof(float) * bias_data.size())));
}
const std::array<int32_t, 4> input_shape{
{BatchSize(), InputHeight(), InputWidth(), InputChannels()}};
const std::array<int32_t, 4> output_shape{
{BatchSize(), OutputHeight(), OutputWidth(), OutputChannels()}};
const std::array<int32_t, 4> filter_shape{
{OutputChannels(), KernelHeight(), KernelWidth(), InputChannels()}};
const std::array<int32_t, 1> bias_shape{{OutputChannels()}};
std::vector<flatbuffers::Offset<tflite::Tensor>> tensors;
if (SparseWeights()) {
// Sparse tensor in TFLite can be in different formats. Here we choose the
// simplest configuration that
// 1. all dimensions are dense,
// 2. in-order traversal, and
// 3. no block configuration.
int dims_count = filter_shape.size();
std::vector<flatbuffers::Offset<DimensionMetadata>> dim_metadata(
dims_count);
std::vector<int> traversal_order(dims_count);
for (int i = 0; i < dims_count; i++) {
traversal_order[i] = i;
dim_metadata[i] = CreateDimensionMetadata(builder, DimensionType_DENSE,
filter_shape[i]);
}
flatbuffers::Offset<SparsityParameters> sparsity_param =
CreateSparsityParameters(builder, builder.CreateVector(traversal_order),
0, builder.CreateVector(dim_metadata));
tensors.emplace_back(CreateTensor(
builder,
builder.CreateVector<int32_t>(filter_shape.data(), filter_shape.size()),
/*type=*/FP16Weights() ? TensorType_FLOAT16 : TensorType_FLOAT32,
/*buffer=*/1, /*name=*/0, /*quantization=*/0,
/*is_variable=*/false, /*sparsity=*/sparsity_param));
}
if (FP16Weights()) {
tensors.emplace_back(CreateTensor(
builder,
builder.CreateVector<int32_t>(filter_shape.data(), filter_shape.size()),
TensorType_FLOAT16, /*buffer=*/SparseWeights() ? 0 : 1));
tensors.emplace_back(CreateTensor(
builder,
builder.CreateVector<int32_t>(bias_shape.data(), bias_shape.size()),
TensorType_FLOAT16, /*buffer=*/2));
}
tensors.emplace_back(CreateTensor(
builder,
builder.CreateVector<int32_t>(input_shape.data(), input_shape.size()),
TensorType_FLOAT32));
tensors.emplace_back(CreateTensor(
builder,
builder.CreateVector<int32_t>(filter_shape.data(), filter_shape.size()),
TensorType_FLOAT32, /*buffer=*/FP16Weights() || SparseWeights() ? 0 : 1));
tensors.emplace_back(CreateTensor(
builder,
builder.CreateVector<int32_t>(bias_shape.data(), bias_shape.size()),
TensorType_FLOAT32, /*buffer=*/FP16Weights() ? 0 : 2));
tensors.emplace_back(CreateTensor(
builder,
builder.CreateVector<int32_t>(output_shape.data(), output_shape.size()),
TensorType_FLOAT32));
const std::array<int32_t, 3> op_inputs{
{static_cast<int>(tensors.size()) - 4,
static_cast<int>(tensors.size()) - 3,
static_cast<int>(tensors.size()) - 2}};
const std::array<int32_t, 1> op_outputs{
{static_cast<int>(tensors.size()) - 1}};
flatbuffers::Offset<Conv2DOptions> conv2d_options =
CreateConv2DOptions(builder, Padding(), StrideWidth(), StrideHeight(),
Activation(), DilationWidth(), DilationHeight());
operators.emplace_back(CreateOperator(
builder, /*opcode_index=*/0,
builder.CreateVector<int32_t>(op_inputs.data(), op_inputs.size()),
builder.CreateVector<int32_t>(op_outputs.data(), op_outputs.size()),
BuiltinOptions_Conv2DOptions, conv2d_options.Union()));
const std::array<int32_t, 1> subgraph_inputs{
{static_cast<int>(tensors.size()) - 4}};
const std::array<int32_t, 1> subgraph_outputs{
{static_cast<int>(tensors.size()) - 1}};
flatbuffers::Offset<SubGraph> subgraph = CreateSubGraph(
builder, builder.CreateVector(tensors.data(), tensors.size()),
builder.CreateVector<int32_t>(subgraph_inputs.data(),
subgraph_inputs.size()),
builder.CreateVector<int32_t>(subgraph_outputs.data(),
subgraph_outputs.size()),
builder.CreateVector(operators.data(), operators.size()));
flatbuffers::Offset<flatbuffers::String> description =
builder.CreateString("Conv2D model");
flatbuffers::Offset<Model> model_buffer = CreateModel(
builder, TFLITE_SCHEMA_VERSION,
builder.CreateVector(operator_codes.data(), operator_codes.size()),
builder.CreateVector(&subgraph, 1), description,
builder.CreateVector(buffers.data(), buffers.size()));
builder.Finish(model_buffer);
return std::vector<char>(builder.GetBufferPointer(),
builder.GetBufferPointer() + builder.GetSize());
}
} // namespace xnnpack
} // namespace tflite