You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

331 lines
14 KiB

/*
* Copyright (C) 2017 The Android Open Source Project
*
* 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 <android-base/scopeguard.h>
#include <gtest/gtest.h>
#include "TestNeuralNetworksWrapper.h"
using namespace android::nn::test_wrapper;
namespace {
typedef float Matrix3x4[3][4];
typedef float Matrix4[4];
const int32_t kNoActivation = ANEURALNETWORKS_FUSED_NONE;
class TrivialTest : public ::testing::Test {
protected:
virtual void SetUp() {}
#if defined(__ANDROID__)
void testAddTwoWithHardwareBufferInput(uint64_t additionalAhwbUsage);
#endif
const Matrix3x4 matrix1 = {{1.f, 2.f, 3.f, 4.f}, {5.f, 6.f, 7.f, 8.f}, {9.f, 10.f, 11.f, 12.f}};
const Matrix3x4 matrix2 = {{100.f, 200.f, 300.f, 400.f},
{500.f, 600.f, 700.f, 800.f},
{900.f, 1000.f, 1100.f, 1200.f}};
const Matrix4 matrix2b = {100.f, 200.f, 300.f, 400.f};
const Matrix3x4 matrix3 = {
{20.f, 30.f, 40.f, 50.f}, {21.f, 22.f, 23.f, 24.f}, {31.f, 32.f, 33.f, 34.f}};
const Matrix3x4 expected2 = {{101.f, 202.f, 303.f, 404.f},
{505.f, 606.f, 707.f, 808.f},
{909.f, 1010.f, 1111.f, 1212.f}};
const Matrix3x4 expected2b = {{101.f, 202.f, 303.f, 404.f},
{105.f, 206.f, 307.f, 408.f},
{109.f, 210.f, 311.f, 412.f}};
const Matrix3x4 expected2c = {{100.f, 400.f, 900.f, 1600.f},
{500.f, 1200.f, 2100.f, 3200.f},
{900.f, 2000.f, 3300.f, 4800.f}};
const Matrix3x4 expected3 = {{121.f, 232.f, 343.f, 454.f},
{526.f, 628.f, 730.f, 832.f},
{940.f, 1042.f, 1144.f, 1246.f}};
const Matrix3x4 expected3b = {
{22.f, 34.f, 46.f, 58.f}, {31.f, 34.f, 37.f, 40.f}, {49.f, 52.f, 55.f, 58.f}};
};
// Create a model that can add two tensors using a one node graph.
void CreateAddTwoTensorModel(Model* model) {
OperandType matrixType(Type::TENSOR_FLOAT32, {3, 4});
OperandType scalarType(Type::INT32, {});
auto a = model->addOperand(&matrixType);
auto b = model->addOperand(&matrixType);
auto c = model->addOperand(&matrixType);
auto d = model->addConstantOperand(&scalarType, kNoActivation);
model->addOperation(ANEURALNETWORKS_ADD, {a, b, d}, {c});
model->identifyInputsAndOutputs({a, b}, {c});
ASSERT_TRUE(model->isValid());
model->finish();
}
// Create a model that can add three tensors using a two node graph,
// with one tensor set as part of the model.
void CreateAddThreeTensorModel(Model* model, const Matrix3x4 bias) {
OperandType matrixType(Type::TENSOR_FLOAT32, {3, 4});
OperandType scalarType(Type::INT32, {});
auto a = model->addOperand(&matrixType);
auto b = model->addOperand(&matrixType);
auto c = model->addOperand(&matrixType);
auto d = model->addOperand(&matrixType);
auto e = model->addOperand(&matrixType);
auto f = model->addConstantOperand(&scalarType, kNoActivation);
model->setOperandValue(e, bias, sizeof(Matrix3x4));
model->addOperation(ANEURALNETWORKS_ADD, {a, c, f}, {b});
model->addOperation(ANEURALNETWORKS_ADD, {b, e, f}, {d});
model->identifyInputsAndOutputs({c, a}, {d});
ASSERT_TRUE(model->isValid());
model->finish();
}
// Check that the values are the same. This works only if dealing with integer
// value, otherwise we should accept values that are similar if not exact.
int CompareMatrices(const Matrix3x4& expected, const Matrix3x4& actual) {
int errors = 0;
for (int i = 0; i < 3; i++) {
for (int j = 0; j < 4; j++) {
if (expected[i][j] != actual[i][j]) {
printf("expected[%d][%d] != actual[%d][%d], %f != %f\n", i, j, i, j,
static_cast<double>(expected[i][j]), static_cast<double>(actual[i][j]));
errors++;
}
}
}
return errors;
}
TEST_F(TrivialTest, AddTwo) {
Model modelAdd2;
CreateAddTwoTensorModel(&modelAdd2);
// Test the one node model.
Matrix3x4 actual;
memset(&actual, 0, sizeof(actual));
Compilation compilation(&modelAdd2);
compilation.finish();
Execution execution(&compilation);
ASSERT_EQ(execution.setInput(0, matrix1, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution.setInput(1, matrix2, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution.setOutput(0, actual, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution.compute(), Result::NO_ERROR);
ASSERT_EQ(CompareMatrices(expected2, actual), 0);
}
// Hardware buffers are an Android concept, which aren't necessarily
// available on other platforms such as ChromeOS, which also build NNAPI.
#if defined(__ANDROID__)
void TrivialTest::testAddTwoWithHardwareBufferInput(uint64_t additionalAhwbUsage) {
Model modelAdd2;
CreateAddTwoTensorModel(&modelAdd2);
const uint64_t cpuUsage =
AHARDWAREBUFFER_USAGE_CPU_READ_OFTEN | AHARDWAREBUFFER_USAGE_CPU_WRITE_OFTEN;
AHardwareBuffer_Desc desc{
.width = sizeof(matrix1),
.height = 1,
.layers = 1,
.format = AHARDWAREBUFFER_FORMAT_BLOB,
.usage = cpuUsage | additionalAhwbUsage,
};
AHardwareBuffer* matrix1Buffer = nullptr;
ASSERT_EQ(AHardwareBuffer_allocate(&desc, &matrix1Buffer), 0);
auto allocateGuard = android::base::make_scope_guard(
[matrix1Buffer]() { AHardwareBuffer_release(matrix1Buffer); });
Memory matrix1Memory(matrix1Buffer);
ASSERT_TRUE(matrix1Memory.isValid());
// Test the one node model.
Matrix3x4 actual;
memset(&actual, 0, sizeof(actual));
Compilation compilation(&modelAdd2);
compilation.finish();
Execution execution(&compilation);
ASSERT_EQ(execution.setInputFromMemory(0, &matrix1Memory, 0, sizeof(Matrix3x4)),
Result::NO_ERROR);
ASSERT_EQ(execution.setInput(1, matrix2, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution.setOutput(0, actual, sizeof(Matrix3x4)), Result::NO_ERROR);
// Set the value for matrix1Buffer.
void* bufferPtr = nullptr;
ASSERT_EQ(AHardwareBuffer_lock(matrix1Buffer, cpuUsage, -1, NULL, &bufferPtr), 0);
memcpy((uint8_t*)bufferPtr, matrix1, sizeof(matrix1));
int synFenceFd = -1;
ASSERT_EQ(AHardwareBuffer_unlock(matrix1Buffer, &synFenceFd), 0);
if (synFenceFd > 0) {
// If valid sync fence is return by AHardwareBuffer_unlock, use
// ANeuralNetworksExecution_startComputeWithDependencies
ANeuralNetworksEvent* eventBufferUnlock;
ANeuralNetworksEvent* eventToSignal;
ASSERT_EQ(ANeuralNetworksEvent_createFromSyncFenceFd(synFenceFd, &eventBufferUnlock),
ANEURALNETWORKS_NO_ERROR);
close(synFenceFd);
ANeuralNetworksExecution* executionHandle = execution.getHandle();
ASSERT_EQ(ANeuralNetworksExecution_startComputeWithDependencies(
executionHandle, &eventBufferUnlock, 1, 0, &eventToSignal),
ANEURALNETWORKS_NO_ERROR);
ASSERT_EQ(ANeuralNetworksEvent_wait(eventToSignal), ANEURALNETWORKS_NO_ERROR);
ANeuralNetworksEvent_free(eventBufferUnlock);
ANeuralNetworksEvent_free(eventToSignal);
} else {
ASSERT_EQ(execution.compute(), Result::NO_ERROR);
}
ASSERT_EQ(CompareMatrices(expected2, actual), 0);
}
TEST_F(TrivialTest, AddTwoWithHardwareBufferInput) {
testAddTwoWithHardwareBufferInput(/* no additional usage */ 0u);
}
TEST_F(TrivialTest, AddTwoWithHardwareBufferInputWithGPUUsage) {
testAddTwoWithHardwareBufferInput(AHARDWAREBUFFER_USAGE_GPU_DATA_BUFFER);
}
#endif
TEST_F(TrivialTest, AddThree) {
Model modelAdd3;
CreateAddThreeTensorModel(&modelAdd3, matrix3);
// Test the three node model.
Matrix3x4 actual;
memset(&actual, 0, sizeof(actual));
Compilation compilation2(&modelAdd3);
compilation2.finish();
Execution execution2(&compilation2);
ASSERT_EQ(execution2.setInput(0, matrix1, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution2.setInput(1, matrix2, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution2.setOutput(0, actual, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution2.compute(), Result::NO_ERROR);
ASSERT_EQ(CompareMatrices(expected3, actual), 0);
// Test it a second time to make sure the model is reusable.
memset(&actual, 0, sizeof(actual));
Compilation compilation3(&modelAdd3);
compilation3.finish();
Execution execution3(&compilation3);
ASSERT_EQ(execution3.setInput(0, matrix1, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution3.setInput(1, matrix1, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution3.setOutput(0, actual, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution3.compute(), Result::NO_ERROR);
ASSERT_EQ(CompareMatrices(expected3b, actual), 0);
}
TEST_F(TrivialTest, FencedAddThree) {
Model modelAdd3;
CreateAddThreeTensorModel(&modelAdd3, matrix3);
Compilation compilation(&modelAdd3);
compilation.finish();
Matrix3x4 output1, output2;
memset(&output1, 0, sizeof(output1));
memset(&output2, 0, sizeof(output2));
// Start the first execution
Execution execution1(&compilation);
ASSERT_EQ(execution1.setInput(0, matrix1, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution1.setInput(1, matrix2, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution1.setOutput(0, output1, sizeof(Matrix3x4)), Result::NO_ERROR);
ANeuralNetworksEvent* event1;
ANeuralNetworksExecution* execution1_handle = execution1.getHandle();
ASSERT_EQ(ANeuralNetworksExecution_startComputeWithDependencies(execution1_handle, nullptr, 0,
0, &event1),
ANEURALNETWORKS_NO_ERROR);
// Start the second execution which will wait for the first one.
Execution execution2(&compilation);
ASSERT_EQ(execution2.setInput(0, matrix1, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution2.setInput(1, matrix1, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution2.setOutput(0, output2, sizeof(Matrix3x4)), Result::NO_ERROR);
ANeuralNetworksEvent* event2;
ANeuralNetworksExecution* execution2_handle = execution2.getHandle();
ASSERT_EQ(ANeuralNetworksExecution_startComputeWithDependencies(execution2_handle, &event1, 1,
0, &event2),
ANEURALNETWORKS_NO_ERROR);
// Wait for the second event.
ASSERT_EQ(ANeuralNetworksEvent_wait(event2), ANEURALNETWORKS_NO_ERROR);
// Check the results for both executions.
ASSERT_EQ(CompareMatrices(expected3, output1), 0);
ASSERT_EQ(CompareMatrices(expected3b, output2), 0);
// Free the event objects
ANeuralNetworksEvent_free(event1);
ANeuralNetworksEvent_free(event2);
}
TEST_F(TrivialTest, BroadcastAddTwo) {
Model modelBroadcastAdd2;
OperandType scalarType(Type::INT32, {});
auto activation = modelBroadcastAdd2.addConstantOperand(&scalarType, kNoActivation);
OperandType matrixType(Type::TENSOR_FLOAT32, {1, 1, 3, 4});
OperandType matrixType2(Type::TENSOR_FLOAT32, {4});
auto a = modelBroadcastAdd2.addOperand(&matrixType);
auto b = modelBroadcastAdd2.addOperand(&matrixType2);
auto c = modelBroadcastAdd2.addOperand(&matrixType);
modelBroadcastAdd2.addOperation(ANEURALNETWORKS_ADD, {a, b, activation}, {c});
modelBroadcastAdd2.identifyInputsAndOutputs({a, b}, {c});
ASSERT_TRUE(modelBroadcastAdd2.isValid());
modelBroadcastAdd2.finish();
// Test the one node model.
Matrix3x4 actual;
memset(&actual, 0, sizeof(actual));
Compilation compilation(&modelBroadcastAdd2);
compilation.finish();
Execution execution(&compilation);
ASSERT_EQ(execution.setInput(0, matrix1, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution.setInput(1, matrix2b, sizeof(Matrix4)), Result::NO_ERROR);
ASSERT_EQ(execution.setOutput(0, actual, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution.compute(), Result::NO_ERROR);
ASSERT_EQ(CompareMatrices(expected2b, actual), 0);
}
TEST_F(TrivialTest, BroadcastMulTwo) {
Model modelBroadcastMul2;
OperandType scalarType(Type::INT32, {});
auto activation = modelBroadcastMul2.addConstantOperand(&scalarType, kNoActivation);
OperandType matrixType(Type::TENSOR_FLOAT32, {1, 1, 3, 4});
OperandType matrixType2(Type::TENSOR_FLOAT32, {4});
auto a = modelBroadcastMul2.addOperand(&matrixType);
auto b = modelBroadcastMul2.addOperand(&matrixType2);
auto c = modelBroadcastMul2.addOperand(&matrixType);
modelBroadcastMul2.addOperation(ANEURALNETWORKS_MUL, {a, b, activation}, {c});
modelBroadcastMul2.identifyInputsAndOutputs({a, b}, {c});
ASSERT_TRUE(modelBroadcastMul2.isValid());
modelBroadcastMul2.finish();
// Test the one node model.
Matrix3x4 actual;
memset(&actual, 0, sizeof(actual));
Compilation compilation(&modelBroadcastMul2);
compilation.finish();
Execution execution(&compilation);
ASSERT_EQ(execution.setInput(0, matrix1, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution.setInput(1, matrix2b, sizeof(Matrix4)), Result::NO_ERROR);
ASSERT_EQ(execution.setOutput(0, actual, sizeof(Matrix3x4)), Result::NO_ERROR);
ASSERT_EQ(execution.compute(), Result::NO_ERROR);
ASSERT_EQ(CompareMatrices(expected2c, actual), 0);
}
} // end namespace