TheRock/rocm-libraries/projects/miopen/test/gtest/tensor_api.cpp

/*******************************************************************************
 *
 * MIT License
 *
 * Copyright (c) 2023 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to deal
 * in the Software without restriction, including without limitation the rights
 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in all
 * copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 *******************************************************************************/

// Test Suite for tensor API

#include <gtest/gtest.h>
#include <miopen/miopen.h>

#include <iomanip>
#include <optional>
#include <sstream>
#include <string>
#include <vector>

// Compiler uses undefined behavior sanitizer
// -fsanitize=enum (or -fsanitize=undefined)
#if(defined(__clang__) || defined(__GNUG__)) && !defined(NDEBUG)
#define UBSAN_ENABLED 1
#else
#define UBSAN_ENABLED 0
#endif

// We use out-of-range values for miopenDataType_t and miopenTensorLayout_t
#define USE_OUT_OF_RANGE_ENUM (UBSAN_ENABLED == 0)

// miopenLastDataType must be changed if new data types are added
#define miopenFirstDataType miopenHalf
#define miopenLastDataType miopenInt64

// miopenLastTensorLayout must be changed if new layouts are added
#define miopenFirstTensorLayout miopenTensorNCHW
#define miopenLastTensorLayout miopenTensorNDHWC

namespace {

using testDataType_t     = int;
using testTensorLayout_t = int;

enum class TestStatus
{
    Failed,
    Passed,
    Skipped
};

struct TensorParams
{
    testDataType_t dataType;
    std::optional<testTensorLayout_t> tensorLayout;
    int nbDims;
    int* dimsA;
    int* stridesA;

    friend std::ostream& operator<<(std::ostream& os, const TensorParams& tp)
    {
        os << "(";
        os << "type:" << static_cast<int>(tp.dataType) << ",";
        os << "layout:";
        if(tp.tensorLayout)
            os << static_cast<int>(tp.tensorLayout.value());
        else
            os << "none";
        os << ",";
        os << "ndims:" << tp.nbDims;
        if(tp.nbDims > 0)
        {
            os << ",";
            if(tp.dimsA == nullptr)
            {
                os << "dimsA:none";
            }
            else
            {
                for(unsigned i = 0; i < tp.nbDims; i++)
                {
                    if(i != 0)
                        os << "x";
                    os << tp.dimsA[i];
                }
            }
            os << ",";
            if(tp.stridesA == nullptr)
            {
                os << "stridesA:none";
            }
            else
            {
                for(unsigned i = 0; i < tp.nbDims; i++)
                {
                    if(i != 0)
                        os << "x";
                    os << tp.stridesA[i];
                }
            }
        }
        os << ")";
        return os;
    }
};

struct TestConfig
{
    bool null_tensor_descriptor;
    TensorParams params;
    bool valid;
    bool skip;

    friend std::ostream& operator<<(std::ostream& os, const TestConfig& tc)
    {
        os << "(";
        os << "skip:" << tc.skip << ",";
        os << "valid:" << tc.valid << ",";
        os << "null_td:" << tc.null_tensor_descriptor << ",";
        os << tc.params;
        os << ")";
        return os;
    }
};

template <class L1, class L2>
bool CompareLengths(const L1 l1, const L2 l2, int size)
{
    for(int i = 0; i < size; i++)
    {
        if(l1[i] != l2[i])
            return false;
    }
    return true;
}

// Set tensor descriptor
TestStatus Set4dTensorDescriptor(miopenTensorDescriptor_t tensorDesc,
                                 const TensorParams& params,
                                 bool check_skip)
{
    if(params.tensorLayout && (params.tensorLayout != miopenTensorNCHW))
        return TestStatus::Skipped;
    if(params.nbDims != 4 || params.dimsA == nullptr || params.stridesA != nullptr)
        return TestStatus::Skipped;

    if(check_skip)
        return TestStatus::Passed;

    miopenStatus_t status =
        miopenSet4dTensorDescriptor(tensorDesc,
                                    static_cast<miopenDataType_t>(params.dataType),
                                    params.dimsA[0],
                                    params.dimsA[1],
                                    params.dimsA[2],
                                    params.dimsA[3]);
    if(status == miopenStatusSuccess)
        return TestStatus::Passed;

    return TestStatus::Failed;
}

TestStatus SetNdTensorDescriptorWithLayout(miopenTensorDescriptor_t tensorDesc,
                                           const TensorParams& params,
                                           bool check_skip)
{
    if(!params.tensorLayout || params.stridesA != nullptr)
        return TestStatus::Skipped;

    if(check_skip)
        return TestStatus::Passed;

    miopenStatus_t status = miopenSetNdTensorDescriptorWithLayout(
        tensorDesc,
        static_cast<miopenDataType_t>(params.dataType),
        static_cast<miopenTensorLayout_t>(params.tensorLayout.value()),
        params.dimsA,
        params.nbDims);
    if(status == miopenStatusSuccess)
        return TestStatus::Passed;

    return TestStatus::Failed;
}

TestStatus Set4dTensorDescriptorEx(miopenTensorDescriptor_t tensorDesc,
                                   const TensorParams& params,
                                   bool check_skip)
{
    if(params.tensorLayout || params.nbDims != 4 || params.dimsA == nullptr ||
       params.stridesA == nullptr)
    {
        return TestStatus::Skipped;
    }

    if(check_skip)
        return TestStatus::Passed;

    miopenStatus_t status =
        miopenSet4dTensorDescriptorEx(tensorDesc,
                                      static_cast<miopenDataType_t>(params.dataType),
                                      params.dimsA[0],
                                      params.dimsA[1],
                                      params.dimsA[2],
                                      params.dimsA[3],
                                      params.stridesA[0],
                                      params.stridesA[1],
                                      params.stridesA[2],
                                      params.stridesA[3]);
    if(status == miopenStatusSuccess)
        return TestStatus::Passed;

    return TestStatus::Failed;
}

TestStatus SetTensorDescriptor(miopenTensorDescriptor_t tensorDesc,
                               const TensorParams& params,
                               bool check_skip)
{
    if(params.tensorLayout)
        return TestStatus::Skipped;

    if(check_skip)
        return TestStatus::Passed;

    miopenStatus_t status =
        miopenSetTensorDescriptor(tensorDesc,
                                  static_cast<miopenDataType_t>(params.dataType),
                                  params.nbDims,
                                  params.dimsA,
                                  params.stridesA);
    if(status == miopenStatusSuccess)
        return TestStatus::Passed;

    return TestStatus::Failed;
}

TestStatus SetTensorDescriptorV2(miopenTensorDescriptor_t tensorDesc,
                                 const TensorParams& params,
                                 bool check_skip)
{
    if(params.tensorLayout)
        return TestStatus::Skipped;

    if(check_skip)
        return TestStatus::Passed;

    size_t* dimsA    = nullptr;
    size_t* stridesA = nullptr;
    std::vector<std::size_t> dims;
    std::vector<std::size_t> strides;

    if(params.nbDims > 0)
    {
        if(params.dimsA != nullptr)
        {
            dims  = std::vector<std::size_t>(params.dimsA, params.dimsA + params.nbDims);
            dimsA = dims.data();
        }
        if(params.stridesA != nullptr)
        {
            strides  = std::vector<std::size_t>(params.stridesA, params.stridesA + params.nbDims);
            stridesA = strides.data();
        }
    }

    miopenStatus_t status = miopenSetTensorDescriptorV2(
        tensorDesc, static_cast<miopenDataType_t>(params.dataType), params.nbDims, dimsA, stridesA);
    if(status == miopenStatusSuccess)
        return TestStatus::Passed;

    return TestStatus::Failed;
}

const auto set_tensor_descr_funcs = {Set4dTensorDescriptor,
                                     SetNdTensorDescriptorWithLayout,
                                     Set4dTensorDescriptorEx,
                                     SetTensorDescriptor,
                                     SetTensorDescriptorV2};

// Get tensor descriptor
TestStatus Get4dTensorDescriptor(miopenTensorDescriptor_t tensorDesc, const TensorParams& params)
{
    if(params.tensorLayout || params.nbDims != 4)
        return TestStatus::Skipped;

    if(params.dimsA == nullptr)
        return TestStatus::Failed; // internal error

    miopenStatus_t status;
    miopenDataType_t dataType;
    int dims[4], strides[4];

    status = miopenGet4dTensorDescriptor(tensorDesc,
                                         &dataType,
                                         dims,
                                         dims + 1,
                                         dims + 2,
                                         dims + 3,
                                         strides,
                                         strides + 1,
                                         strides + 2,
                                         strides + 3);
    if(status != miopenStatusSuccess)
        return TestStatus::Failed;

    if(params.dataType != dataType || !CompareLengths(params.dimsA, dims, 4) ||
       (params.stridesA != nullptr && !CompareLengths(params.stridesA, strides, 4)))
    {
        return TestStatus::Failed;
    }

    return TestStatus::Passed;
}

TestStatus GetTensorDescriptor(miopenTensorDescriptor_t tensorDesc, const TensorParams& params)
{
    if(params.tensorLayout)
    {
        if((params.tensorLayout != miopenTensorNCHW && params.tensorLayout != miopenTensorNCDHW) ||
           params.stridesA != nullptr)
        {
            return TestStatus::Skipped;
        }
    }

    if(params.dimsA == nullptr)
        return TestStatus::Failed; // internal error

    miopenStatus_t status;
    int size;

    status = miopenGetTensorDescriptorSize(tensorDesc, &size);
    if(status != miopenStatusSuccess || size < 0 || size != params.nbDims)
        return TestStatus::Failed;

    miopenDataType_t dataType;
    std::vector<int> dims(size);
    std::vector<int> strides(size);

    status = miopenGetTensorDescriptor(tensorDesc, &dataType, dims.data(), strides.data());
    if(status != miopenStatusSuccess)
        return TestStatus::Failed;

    if(params.dataType != dataType || !CompareLengths(params.dimsA, dims, size) ||
       (params.stridesA != nullptr && !CompareLengths(params.stridesA, strides, size)))
        return TestStatus::Failed;

    return TestStatus::Passed;
}

const auto get_tensor_descr_funcs = {Get4dTensorDescriptor, GetTensorDescriptor};

class TestSetTensor : public ::testing::TestWithParam<TestConfig>
{
protected:
    static int GetNumDimsForLayout(testTensorLayout_t layout)
    {
        int ndims = 0;
        // clang-format off
        switch(layout)
        {
        case miopenTensorNCHW:
        case miopenTensorNHWC:
        case miopenTensorCHWN:
        case miopenTensorNCHWc4:
        case miopenTensorNCHWc8:
        case miopenTensorCHWNc4:
        case miopenTensorCHWNc8:
            ndims = 4;
            break;
        case miopenTensorNCDHW:
        case miopenTensorNDHWC:
            ndims = 5;
            break;
        }
        // clang-format on
        return ndims;
    }

    static void GenerateValidConfigs(std::vector<TestConfig>& configs)
    {
        static int dims[]    = {4, 4, 16, 9, 16};
        static int strides[] = {9216, 2304, 144, 16, 1};
        static_assert(sizeof(dims) == sizeof(strides));
        const auto max_ndims = sizeof(dims) / sizeof(dims[0]);

        // clang-format off
        for(testDataType_t datatype = miopenFirstDataType; datatype <= miopenLastDataType; datatype++)
        // clang-format on
        {
            if(datatype == 4)
                continue; // miopenInt8x4

            // clang-format off
            for(testTensorLayout_t layout = miopenFirstTensorLayout; layout <= miopenLastTensorLayout; layout++)
            // clang-format on
            {
                int ndims               = GetNumDimsForLayout(layout);
                const TestConfig config = {
                    false,
                    {datatype, layout, ndims, dims + (max_ndims - ndims), nullptr},
                    true,
                    false};
                configs.push_back(config);
            }

            for(int i = 0; i < 2; i++)
            {
                const bool use_strides = (i == 1);
                for(int ndims = 1; ndims <= max_ndims; ndims++)
                {
                    const TestConfig config = {
                        false,
                        {datatype,
                         std::nullopt,
                         ndims,
                         dims + (max_ndims - ndims),
                         use_strides ? (strides + (max_ndims - ndims)) : nullptr},
                        true,
                        false};
                    configs.push_back(config);
                }
            }
        }
    }

    static std::vector<TestConfig> GenerateValidConfigs()
    {
        std::vector<TestConfig> configs;
        GenerateValidConfigs(configs);
        return configs;
    }

    static std::vector<testDataType_t> GetWrongDataTypes()
    {
        std::vector<testDataType_t> wrong_datatypes = {
            static_cast<testDataType_t>(miopenFirstDataType) - 1,
            static_cast<testDataType_t>(miopenLastDataType) + 1,
            /*miopenInt8x4*/ 4};
        return wrong_datatypes;
    }

    static std::vector<testTensorLayout_t> GetWrongLayouts(int num_dims, bool use_strides)
    {
        std::vector<testTensorLayout_t> wrong_layouts = {
            static_cast<testTensorLayout_t>(miopenFirstTensorLayout) - 1,
            static_cast<testTensorLayout_t>(miopenLastTensorLayout) + 1};
        std::vector<testTensorLayout_t> layouts_4d = {miopenTensorNCHW,
                                                      miopenTensorNHWC,
                                                      miopenTensorCHWN,
                                                      miopenTensorNCHWc4,
                                                      miopenTensorNCHWc8,
                                                      miopenTensorCHWNc4,
                                                      miopenTensorCHWNc8};
        std::vector<testTensorLayout_t> layouts_5d = {miopenTensorNCDHW, miopenTensorNDHWC};
        if(use_strides)
        {
            wrong_layouts.insert(wrong_layouts.end(), layouts_4d.cbegin(), layouts_4d.cend());
            wrong_layouts.insert(wrong_layouts.end(), layouts_5d.cbegin(), layouts_5d.cend());
        }
        else
        {
            if(num_dims == 4)
                wrong_layouts.insert(wrong_layouts.end(), layouts_5d.cbegin(), layouts_5d.cend());
            else if(num_dims == 5)
                wrong_layouts.insert(wrong_layouts.end(), layouts_4d.cbegin(), layouts_4d.cend());
        }
        return wrong_layouts;
    }

    static std::vector<int> GetWrongNumDims(std::optional<testTensorLayout_t> layout)
    {
        std::vector<int> wrong_ndims = {-1, 0};
        if(!layout)
            return wrong_ndims;

        auto num_dims = GetNumDimsForLayout(layout.value());
        wrong_ndims.push_back(num_dims - 1);
        wrong_ndims.push_back(num_dims + 1);

        return wrong_ndims;
    }

    static void GenerateWrongConfigs(const TestConfig& valid_config,
                                     std::vector<TestConfig>& wrong_configs)
    {
        const auto wrong_datatypes = GetWrongDataTypes();
        const auto wrong_layouts =
            GetWrongLayouts(valid_config.params.nbDims, valid_config.params.stridesA != nullptr);
        const auto wrong_ndims     = GetWrongNumDims(valid_config.params.tensorLayout);
        static int wrong_dims[][8] = {{0, 0, 0, 0, 0, 0, 0, 0}, {-1, -1, -1, -1, -1, -1, -1, -1}};

        // tensorDesc = nullptr
        {
            auto config                   = valid_config;
            config.null_tensor_descriptor = true;
            config.valid                  = false;
            wrong_configs.push_back(config);
        }

        // wrong data type
        for(auto datatype : wrong_datatypes)
        {
            auto config            = valid_config;
            config.params.dataType = datatype;
            config.valid           = false;
#if !USE_OUT_OF_RANGE_ENUM
            if(config.params.dataType < miopenFirstDataType ||
               config.params.dataType > miopenLastDataType)
            {
                config.skip = true;
            }
#endif
            wrong_configs.push_back(config);
        }

        // wrong layout
        for(auto layout : wrong_layouts)
        {
            auto config                = valid_config;
            config.params.tensorLayout = layout;
            config.valid               = false;
#if !USE_OUT_OF_RANGE_ENUM
            if(config.params.tensorLayout < miopenFirstTensorLayout ||
               config.params.tensorLayout > miopenLastTensorLayout)
            {
                config.skip = true;
            }
#endif
            wrong_configs.push_back(config);
        }

#if defined(__SANITIZE_ADDRESS__) || (defined(__has_feature) && __has_feature(address_sanitizer))
        // Skip wrong dimension tests when AddressSanitizer is enabled as they trigger buffer
        // overflows before reaching actual API validation, rendering the edge case tests useless.
#else
        // wrong number of dimensions
        for(auto ndims : wrong_ndims)
        {
            auto config          = valid_config;
            config.params.nbDims = ndims;
            config.valid         = false;
            wrong_configs.push_back(config);
        }
#endif

        // dimsA = nullptr
        {
            auto config         = valid_config;
            config.params.dimsA = nullptr;
            config.valid        = false;
            wrong_configs.push_back(config);
        }

        // wrong dimensions
        for(auto dims : wrong_dims)
        {
            auto config         = valid_config;
            config.params.dimsA = dims;
            config.valid        = false;
            wrong_configs.push_back(config);
        }

        if(valid_config.params.stridesA != nullptr)
        {
            // wrong strides
            for(auto strides : wrong_dims)
            {
                auto config            = valid_config;
                config.params.stridesA = strides;
                config.valid           = false;
                wrong_configs.push_back(config);
            }
        }
    }

    static std::vector<TestConfig> GenerateWrongConfigs()
    {
        const auto& valid_configs = GetValidConfigs();
        std::vector<TestConfig> wrong_configs;
        for(const auto& valid_config : valid_configs)
            GenerateWrongConfigs(valid_config, wrong_configs);
        return wrong_configs;
    }

    static void RunTest(const TestConfig& config)
    {
        for(const auto set_tensor_descr_func : set_tensor_descr_funcs)
        {
            TestStatus test_status;
            miopenStatus_t status;
            miopenTensorDescriptor_t desc = nullptr;

            test_status = set_tensor_descr_func(nullptr, config.params, true);
            if(test_status == TestStatus::Skipped)
                continue;

            if(!config.null_tensor_descriptor)
            {
                status = miopenCreateTensorDescriptor(&desc);
                ASSERT_EQ(status, miopenStatusSuccess);
            }

            test_status = set_tensor_descr_func(desc, config.params, false);
            if(config.valid)
                ASSERT_NE(test_status, TestStatus::Failed);
            else
                ASSERT_NE(test_status, TestStatus::Passed);

            if(config.valid)
            {
                for(const auto get_tensor_descr_func : get_tensor_descr_funcs)
                {
                    test_status = get_tensor_descr_func(desc, config.params);
                    ASSERT_NE(test_status, TestStatus::Failed);
                }
            }

            if(!config.null_tensor_descriptor)
            {
                status = miopenDestroyTensorDescriptor(desc);
                ASSERT_EQ(status, miopenStatusSuccess);
            }
        }
    }

    static void RunTest()
    {
        const auto& config = GetParam();
        if(config.skip)
            GTEST_SKIP();
        RunTest(config);
    }

public:
    static const std::vector<TestConfig>& GetValidConfigs()
    {
        static const auto configs = GenerateValidConfigs();
        return configs;
    }

    static const std::vector<TestConfig>& GetWrongConfigs()
    {
        static const auto configs = GenerateWrongConfigs();
        return configs;
    }
};

} // namespace

using CPU_ApiTestSetTensorDescriptor_NONE      = TestSetTensor;
using CPU_ApiTestSetWrongTensorDescriptor_NONE = TestSetTensor;

TEST_P(CPU_ApiTestSetTensorDescriptor_NONE, TD) { RunTest(); }

TEST_P(CPU_ApiTestSetWrongTensorDescriptor_NONE, TD) { RunTest(); }

INSTANTIATE_TEST_SUITE_P(Full,
                         CPU_ApiTestSetTensorDescriptor_NONE,
                         testing::ValuesIn(TestSetTensor::GetValidConfigs()));

INSTANTIATE_TEST_SUITE_P(Full,
                         CPU_ApiTestSetWrongTensorDescriptor_NONE,
                         testing::ValuesIn(TestSetTensor::GetWrongConfigs()));