cavis/libnd4j/tests_cpu/layers_tests/PlaygroundTests.cpp

/*******************************************************************************
 * Copyright (c) 2015-2018 Skymind, Inc.
 * Copyright (c) 2019 Konduit K.K.
 *
 * This program and the accompanying materials are made available under the
 * terms of the Apache License, Version 2.0 which is available at
 * https://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.
 *
 * SPDX-License-Identifier: Apache-2.0
 ******************************************************************************/

//
// Created by raver119 on 20.11.17.
//

#include "testlayers.h"
#include <graph/Graph.h>
#include <chrono>
#include <graph/Node.h>
#include <ops/declarable/CustomOperations.h>
#include <graph/profiling/GraphProfilingHelper.h>
#include <loops/type_conversions.h>
#include <helpers/threshold.h>
#include <helpers/MmulHelper.h>
#include <ops/ops.h>
#include <helpers/OmpLaunchHelper.h>
#include <helpers/GradCheck.h>
#include <ops/declarable/helpers/im2col.h>
#include <helpers/Loops.h>
#include <helpers/RandomLauncher.h>
#include <ops/declarable/helpers/convolutions.h>

#include <helpers/BenchmarkHelper.h>
#include <ops/declarable/helpers/scatter.h>
#include <helpers/ConstantShapeHelper.h>
#include <helpers/ConstantTadHelper.h>
#include <array>
#include <performance/benchmarking/FullBenchmarkSuit.h>
#include <performance/benchmarking/LightBenchmarkSuit.h>
#include <random>
#include <ops/declarable/helpers/legacy_helpers.h>
#include <ops/declarable/helpers/addBias.h>
#include <ops/declarable/helpers/axis.h>
#include <ops/declarable/helpers/reductions.h>
#include <helpers/LoopsCoordsHelper.h>
#include <build_info.h>
using namespace sd;
using namespace sd::graph;

class PlaygroundTests : public testing::Test {
public:
    int numIterations = 3;
    int poolSize = 10;

    PlaygroundTests() {
    }
};

TEST_F(PlaygroundTests, test_avx) {
    nd4j_printf("Optimal level: %i; Binary level: %i;\n", ::optimalLevel(), ::binaryLevel());
}

TEST_F(PlaygroundTests, buildver) {
    nd4j_printf("%s\n", buildInfo());
}

TEST_F(PlaygroundTests, test_biasAdd_1) {
    auto x = NDArrayFactory::create<float>('c', {512, 3072});
    auto y = NDArrayFactory::create<float>('c', {3072});

    std::vector<Nd4jLong> values;

    sd::ops::biasadd op;

    for (int e = 0; e < 100; e++) {
        auto timeStart = std::chrono::system_clock::now();

        op.execute({&x, &y}, {&x});

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }

    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);
}


TEST_F(PlaygroundTests, test_bert_full_1) {
#ifdef _RELEASE

    // this test will run ONLY if this model exists
    if (sd::graph::getFileSize("/home/raver119/Downloads/BertFull/model.fb") < 0)
        return;

    auto graph = GraphExecutioner::importFromFlatBuffers("/home/raver119/Downloads/BertFull/model.fb");

    auto t = NDArrayFactory::fromNpyFile("/home/raver119/Downloads/BertFull/in0_IteratorGetNext.npy");
    auto u = NDArrayFactory::fromNpyFile("/home/raver119/Downloads/BertFull/in1_IteratorGetNext_1.npy");
    auto v = NDArrayFactory::fromNpyFile("/home/raver119/Downloads/BertFull/in2_IteratorGetNext_4.npy");
    auto z = NDArrayFactory::fromNpyFile("/home/raver119/Downloads/BertFull/out_loss-Softmax.npy");

    //graph->printOut();

    graph->tagInplaceNodes();

    graph->getVariableSpace()->putVariable(658,0, t);
    graph->getVariableSpace()->putVariable(659,0, u);
    graph->getVariableSpace()->putVariable(660,0, v);

/*
    // validating graph now
    auto status = GraphExecutioner::execute(graph);
    ASSERT_EQ(Status::OK(), status);
    ASSERT_TRUE(graph->getVariableSpace()->hasVariable(1620));

    auto array = graph->getVariableSpace()->getVariable(1620)->getNDArray();
    ASSERT_EQ(z, *array);

*/

    sd::Environment::getInstance().setProfiling(true);
    auto profile = GraphProfilingHelper::profile(graph, 1);

    profile->printOut();

    sd::Environment::getInstance().setProfiling(false);
    delete profile;

/*
    std::vector<Nd4jLong> values;

    for (int e = 0; e < 1; e++) {
        auto timeStart = std::chrono::system_clock::now();

        GraphExecutioner::execute(graph);

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }

    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);
*/
    delete graph;
#endif
}


TEST_F(PlaygroundTests, test_bert_1) {
#ifdef _RELEASE
    // this test will run ONLY if this model exists
    if (sd::graph::getFileSize("/home/raver119/Downloads/Bert_minimal_model/bert_minimal_model.fb") < 0)
        return;

    auto graph = GraphExecutioner::importFromFlatBuffers("/home/raver119/Downloads/Bert_minimal_model/bert_minimal_model.fb");

    auto t = NDArrayFactory::fromNpyFile("/home/raver119/Downloads/Bert_minimal_model/bert_minimal_input_IteratorGetNext.numpy");
    auto u = NDArrayFactory::fromNpyFile("/home/raver119/Downloads/Bert_minimal_model/bert_minimal_input_IteratorGetNext_1.numpy");
    auto v = NDArrayFactory::fromNpyFile("/home/raver119/Downloads/Bert_minimal_model/bert_minimal_input_IteratorGetNext_4.numpy");
    auto z = NDArrayFactory::fromNpyFile("/home/raver119/Downloads/Bert_minimal_model/bert_minimal_model_output.numpy");

    //graph->printOut();

    graph->tagInplaceNodes();

    graph->getVariableSpace()->putVariable(85,0, t);
    graph->getVariableSpace()->putVariable(86,0, u);
    graph->getVariableSpace()->putVariable(87,0, v);

/*
    // validating graph now
    auto status = GraphExecutioner::execute(graph);
    ASSERT_EQ(Status::OK(), status);
    ASSERT_TRUE(graph->getVariableSpace()->hasVariable(198));

    auto array = graph->getVariableSpace()->getVariable(198)->getNDArray();
    ASSERT_EQ(z, *array);

*/
    sd::Environment::getInstance().setProfiling(true);
    auto profile = GraphProfilingHelper::profile(graph, 1);

    profile->printOut();

    sd::Environment::getInstance().setProfiling(false);
    delete profile;

/*
    std::vector<Nd4jLong> values;

    for (int e = 0; e < 1; e++) {
        auto timeStart = std::chrono::system_clock::now();

        GraphExecutioner::execute(graph);

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }

    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);
*/
    delete graph;
#endif
}

TEST_F(PlaygroundTests, test_bert_2) {
#ifdef _RELEASE
    // this test will run ONLY if this model exists
    if (sd::graph::getFileSize("/home/raver119/Downloads/Bert_minimal_model/bert_like_ops.fb") < 0)
        return;

    auto graph = GraphExecutioner::importFromFlatBuffers("/home/raver119/Downloads/Bert_minimal_model/bert_like_ops.fb");

    //graph->printOut();

    graph->tagInplaceNodes();


/*
    // validating graph now
    auto status = GraphExecutioner::execute(graph);
    ASSERT_EQ(Status::OK(), status);
    ASSERT_TRUE(graph->getVariableSpace()->hasVariable(198));

    auto array = graph->getVariableSpace()->getVariable(198)->getNDArray();
    ASSERT_EQ(z, *array);
*/

    sd::Environment::getInstance().setProfiling(true);
    auto profile = GraphProfilingHelper::profile(graph, 1);

    profile->printOut();

    sd::Environment::getInstance().setProfiling(false);
    delete profile;

/*
    std::vector<Nd4jLong> values;

    for (int e = 0; e < 1; e++) {
        auto timeStart = std::chrono::system_clock::now();

        GraphExecutioner::execute(graph);

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }

    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);
*/
    delete graph;
#endif
}


TEST_F(PlaygroundTests, test_one_off_ops_1) {
    auto x = NDArrayFactory::create<float>('c', {4, 128, 768});
    auto y = NDArrayFactory::create<float>('c', {4, 128, 1});
    auto z = x.ulike();

    sd::ops::squaredsubtract op;
    op.execute({&x, &y}, {&z});
}

#if defined(INDEX_REDUCTIONS_BENCH_TESTS)
//temporarly, testing against the original one
void original_argmax(const NDArray& input, std::vector<int>& axis, NDArray& output) {
    sd::ops::helpers::adjustAxis(input.rankOf(), axis);
    input.applyIndexReduce(sd::indexreduce::IndexMax, output, axis);
}

template<typename T>
void fill_random(sd::NDArray& arr) {
    Nd4jLong coords[MAX_RANK] = {};
    std::random_device rd;
    std::mt19937 gen(rd());
    //for floats
    std::uniform_real_distribution<T> dis((T)-10.0, (T)22.9);
    T* x = arr.bufferAsT<T>();
    Nd4jLong* shapeInfo = arr.getShapeInfo();
    Nd4jLong* strides = arr.stridesOf();
    Nd4jLong rank = shapeInfo[0];
    Nd4jLong* bases = &(shapeInfo[1]);
    size_t t = 1;
    for (size_t i = 0; i < rank ; i++) {
        t *= bases[i];
    }
    size_t offset = 0;
    if (arr.ordering() == 'c') {

        for (size_t i = 0; i < t; i++) {
            x[offset] = dis(gen) ;
            offset = sd::inc_coords(bases, strides, coords, offset, rank);
        }

    }
    else {

        for (size_t i = 0; i < t; i++) {
            x[offset] = dis(gen) ;
            offset = sd::inc_coords<false>(bases, strides, coords, offset, rank);
        }

    }
}

void testLegacy(bool random) {
#if 0
    int bases[] = { 3, 2, 4, 5, 7 };
    constexpr int Loop = 1;
#else
    int bases[] = { 8, 32, 64, 32, 64 };
    constexpr int Loop = 10;
#endif
    constexpr int N = 5;

    auto x = NDArrayFactory::create<float>('c', { bases[0], bases[1], bases[2], bases[3], bases[4] });
    if (!random) {
        x.linspace(1);
    }
    else{
        fill_random<float>(x);
     }

#define COMBINATIONS 1
#if COMBINATIONS
//https://www.rosettacode.org/wiki/Combinations#C.2B.2B
for (int k = N; k >= 1; k--) {

    std::string bitmask(k, 1); // K leading 1's
    bitmask.resize(N, 0); // N-K trailing 0's

    do {


        std::vector<int> dimension;

        std::vector<Nd4jLong> output_bases;

        for (int i = 0; i < N; ++i) // [0..N-1] integers
        {
            if (bitmask[i])  dimension.push_back(i);
            else {
                output_bases.push_back(bases[i]);
            }
        }
#else
std::vector<int> dimension = { 0,1,2,3 };
int k = 4;
#endif
auto dim = NDArrayFactory::create<int>(dimension);

#if 1
nd4j_printf("C(N:%d K:%d) \n", N, k);
dim.printIndexedBuffer("Dimension");
for (int xind : dimension) {
    nd4j_printf(" %d ,", bases[xind]);
}
nd4j_printf("%s", "\n");
#endif



std::vector<Nd4jLong> values;
sd::ResultSet result;
for (int e = 0; e < Loop; e++) {
    auto timeStart = std::chrono::system_clock::now();
    NDArray exp = output_bases.size() > 0 ? NDArrayFactory::create<Nd4jLong>('c', output_bases) : NDArrayFactory::create<Nd4jLong>(0);
    original_argmax(x, dimension, exp);
    auto timeEnd = std::chrono::system_clock::now();
    auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
    values.emplace_back(outerTime);
}

std::sort(values.begin(), values.end());

nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);
#if COMBINATIONS

    } while (std::prev_permutation(bitmask.begin(), bitmask.end()));

}
#endif
}

#define DEBUG 1

void testNewReduction(bool random, bool checkCorrectness = false , char order ='c') {
    std::vector<Nd4jLong> arr_dimensions;
#if defined(DEBUG)
    int bases[] = { 3, 2, 3, 3, 5 ,4,7,4,7,7 };
    constexpr int Loop = 1;
    constexpr int N = 10;
#else
    int bases[] = { 8, 32, 64, 32, 64 };
    constexpr int Loop = 10;
    constexpr int N = 5;

#endif

    for (int i = 0; i < N; i++) {
        arr_dimensions.push_back(bases[i]);
    }
    auto x = NDArrayFactory::create<float>(order,arr_dimensions);
    if (!random) {
        x.linspace(1);
    }
    else {
        fill_random<float>(x);
    }

#define COMBINATIONS 1
#if COMBINATIONS
    //https://www.rosettacode.org/wiki/Combinations#C.2B.2B
    for (int k = N; k >= 1; k--) {

        std::string bitmask(k, 1); // K leading 1's
        bitmask.resize(N, 0); // N-K trailing 0's

        do {


            std::vector<int> dimension;

            std::vector<Nd4jLong> output_bases;

            for (int i = 0; i < N; ++i) // [0..N-1] integers
            {
                if (bitmask[i])  dimension.push_back(i);
                else {
                    output_bases.push_back(bases[i]);
                }
            }
#else
    std::vector<int> dimension = { 0,1,2,3 };
    int k = 4;
#endif
    auto dim = NDArrayFactory::create<int>(dimension);

#if 1
    nd4j_printf("C(N:%d K:%d) \n", N, k);
    dim.printIndexedBuffer("Dimension");
    for (int xind : dimension) {
        nd4j_printf(" %d ,", bases[xind]);
    }
    nd4j_printf("%s", "\n");
#endif


    sd::ops::argmax op;
    std::vector<Nd4jLong> values;
    sd::ResultSet result;
    for (int e = 0; e < Loop; e++) {
        auto timeStart = std::chrono::system_clock::now();
        result = op.evaluate({ &x, &dim }, {}, {});
        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }
    auto z = result.at(0);

    if (checkCorrectness) {
        //check for the correctness
        NDArray exp = output_bases.size() > 0 ? NDArrayFactory::create<Nd4jLong>('c', output_bases) : NDArrayFactory::create<Nd4jLong>(0);
        original_argmax(x, dimension, exp);


#if  0// defined(DEBUG)
     x.printIndexedBuffer("X");
    exp.printIndexedBuffer("Expected");
    z->printIndexedBuffer("Z");
#endif

        ASSERT_TRUE(exp.isSameShape(z));
        ASSERT_TRUE(exp.equalsTo(z));
    }
    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);
#if COMBINATIONS

        } while (std::prev_permutation(bitmask.begin(), bitmask.end()));

    }
#endif
}

constexpr bool test_corr = true;
#if !defined(DEBUG)
TEST_F(PlaygroundTests, ArgMaxPerfLinspace) {
    testNewReduction(false, test_corr);
}
#endif

TEST_F(PlaygroundTests, ArgMaxPerfRandom) {
    testNewReduction(true, test_corr);
}

TEST_F(PlaygroundTests, ArgMaxPerfRandomOrderF) {
    testNewReduction(true, test_corr, 'f');
}

#if !defined(DEBUG)
TEST_F(PlaygroundTests, ArgMaxPerfLegacyLinspace) {
    testLegacy(false);
}

TEST_F(PlaygroundTests, ArgMaxPerfLegacyRandom) {
    testLegacy(true);
}

#endif

#endif

/*

TEST_F(PlaygroundTests, test_broadcast_1) {
    int pool = 1000;
    std::vector<NDArray*> aX(pool);
    std::vector<NDArray*> aY(pool);
    std::vector<NDArray*> aZ(pool);

    for (int e = 0; e < pool; e++) {
        aX[e] = NDArrayFactory::create_<float>('c', {512, 3072});
        aY[e] = NDArrayFactory::create_<float>('c', {3072});
        aZ[e] = NDArrayFactory::create_<float>('c', {512, 3072});

        aX[e]->assign(119 * (e+1));
        aY[e]->assign(119 * (e+3));
    }

    std::vector<Nd4jLong> values;
    Context ctx(1);

    sd::ops::biasadd op;

    for (int e = 0; e < 1000; e++) {
        auto x = aX[e < pool ? e : e % pool];
        auto y = aY[e < pool ? e : e % pool];
        auto z = aZ[e < pool ? e : e % pool];

        auto timeStart = std::chrono::system_clock::now();

        //op.execute({x, y}, {z});
        sd::ops::helpers::addBias(ctx, *x, *y, *z, false);

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }

    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);

    for (int e = 0; e < pool; e++) {
        delete aX[e];
        delete aY[e];
        delete aZ[e];
    }
}


/*
TEST_F(PlaygroundTests, test_broadcast_1) {
    int pool = 500;
    std::vector<NDArray*> aX(pool);
    std::vector<NDArray*> aY(pool);
    std::vector<NDArray*> aZ(pool);

    for (int e = 0; e < pool; e++) {
        aX[e] = NDArrayFactory::create_<float>('c', {512, 3072});
        aY[e] = NDArrayFactory::create_<float>('c', {768});
        aZ[e] = NDArrayFactory::create_<float>('c', {512, 3072});

        aX[e]->assign( (e+1) / 119);
        aY[e]->assign( (e+3) / 119);
    }



    std::vector<Nd4jLong> values;

    for (int e = 0; e < 1000; e++) {
        auto x = aX[e < pool ? e : e % pool];
        auto y = aY[e < pool ? e : e % pool];
        auto z = aZ[e < pool ? e : e % pool];

        auto timeStart = std::chrono::system_clock::now();

        //x->applyTrueBroadcast(BroadcastOpsTuple::Multiply(), *y, *z);
        x->applyTransform(transform::Tanh, *z, nullptr);

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }

    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);

    for (int e = 0; e < pool; e++) {
        delete aX[e];
        delete aY[e];
        delete aZ[e];
    }
}

*/
/*

TEST_F(PlaygroundTests, test_s_0) {
    std::vector<std::vector<Nd4jLong>> shapes = {{32, 224, 224, 3}, {32, 56, 56, 64}, {32, 7, 7, 512}};
    std::vector<int> threads = {1, 2, 4, 8, 16};

    for (auto shape: shapes) {
        for (auto t: threads) {
            sd::Environment::getInstance().setMaxMasterThreads(t);

            auto x = NDArrayFactory::create<float>('c', shape);
            auto y = NDArrayFactory::create<float>('c', {shape[3]});
            auto z = x.ulike();

            std::vector<Nd4jLong> values;
            Context ctx(1);
            ctx.setInputArray(0, &x);
            ctx.setInputArray(1, &y);
            ctx.setOutputArray(0, &z);

            sd::ops::biasadd op;


            for (int e = 0; e < 10000; e++) {
                auto timeStart = std::chrono::system_clock::now();

                op.execute(&ctx);
                sd::ops::helpers::addBias(ctx, x, y, z, false);

                auto timeEnd = std::chrono::system_clock::now();
                auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
                values.emplace_back(outerTime);
            }

            std::sort(values.begin(), values.end());

            nd4j_printf("Shape: [%lld, %lld, %lld, %lld]; Threads: [%i]; Time: %lld us;\n", shape[0], shape[1], shape[2], shape[3], t, values[values.size() / 2]);
        }
    }
}

TEST_F(PlaygroundTests, test_s_1) {
    std::vector<std::vector<Nd4jLong>> shapes = {{32, 3, 224, 224}, {32, 64, 56, 56}, {32, 512, 7, 7}};
    std::vector<int> threads = {1, 2, 4, 8, 16};

    for (auto shape: shapes) {
        for (auto t: threads) {
            sd::Environment::getInstance().setMaxMasterThreads(t);

            auto x = NDArrayFactory::create<float>('c', shape);
            auto y = NDArrayFactory::create<float>('c', {shape[1]});
            auto z = x.ulike();

            std::vector<Nd4jLong> values;
            Context ctx(1);
            ctx.setInputArray(0, &x);
            ctx.setInputArray(1, &y);
            ctx.setOutputArray(0, &z);

            sd::ops::biasadd op;


            for (int e = 0; e < 10000; e++) {
                auto timeStart = std::chrono::system_clock::now();

                //op.execute({&x, &y}, {&z}, {true});
                sd::ops::helpers::addBias(ctx, x, y, z, true);

                auto timeEnd = std::chrono::system_clock::now();
                auto outerTime = std::chrono::duration_cast<std::chrono::microseconds>(timeEnd - timeStart).count();
                values.emplace_back(outerTime);
            }

            std::sort(values.begin(), values.end());

            nd4j_printf("Shape: [%lld, %lld, %lld, %lld]; Threads: [%i]; Time: %lld us;\n", shape[0], shape[1], shape[2], shape[3], t, values[values.size() / 2]);
        }
    }
}
*/

/*
TEST_F(PlaygroundTests, test_s_0) {
    auto x = NDArrayFactory::create<float>('c', {32, 112, 112, 16});
    auto y = NDArrayFactory::create<float>('c', {16});
    auto z = x.ulike();

    std::vector<Nd4jLong> values;
    Context ctx(1);
    ctx.setInputArray(0, &x);
    ctx.setInputArray(1, &y);
    ctx.setOutputArray(0, &z);

    sd::ops::biasadd op;


    for (int e = 0; e < 10000; e++) {
        auto timeStart = std::chrono::system_clock::now();

        op.execute(&ctx);

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::microseconds> (timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }

    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);
}
*/
/*
TEST_F(PlaygroundTests, test_s_1) {
    auto x0 = NDArrayFactory::create<float>('c', {32, 7, 7, 176});
    auto x1 = x0.ulike();
    auto x2 = x0.ulike();
    auto x3 = x0.ulike();
    auto x4 = x0.ulike();
    auto x5 = x0.ulike();

    auto y = NDArrayFactory::create<int >(3);
    auto z = NDArrayFactory::create<float>('c', {32, 7, 7, 1056});

    Context ctx(1);
    ctx.setInputArray(0, &x0);
    ctx.setInputArray(1, &x1);
    ctx.setInputArray(2, &x2);
    ctx.setInputArray(3, &x3);
    ctx.setInputArray(4, &x4);
    ctx.setInputArray(5, &x5);

    ctx.setInputArray(6, &y);
    ctx.setOutputArray(0, &z);
    ctx.setBArguments({true});

    std::vector<Nd4jLong> values;

    sd::ops::concat op;
    op.execute(&ctx);

    for (int e = 0; e < 1000; e++) {
        auto timeStart = std::chrono::system_clock::now();

        op.execute(&ctx);

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::microseconds> (timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }


    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld us;\n", values[values.size() / 2]);
}
*/

/*
TEST_F(PlaygroundTests, test_s_1) {
    auto t = ::runLightBenchmarkSuit(true);
    delete[] t;
}

TEST_F(PlaygroundTests, test_s_2) {
    std::atomic<int> s;
    s = 0;
    auto func = PRAGMA_THREADS_FOR {
        s++;
    };

    samediff::Threads::parallel_for(func, 0, 8192, 1, 4);
    std::vector<Nd4jLong> values;

    for (int e = 0; e < 100000; e++) {
        s = 0;

        auto timeStart = std::chrono::system_clock::now();
        //samediff::Threads::parallel_for(func, 0, 8192, 1, 4);
        PRAGMA_OMP_PARALLEL_THREADS(4) {
            s++;
        }

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::nanoseconds> (timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    };
    std::sort(values.begin(), values.end());

    nd4j_printf("Time: %lld;\n", values[values.size() / 2]);
}
 */
/*
TEST_F(PlaygroundTests, test_s_4) {
    std::atomic<float> f;
    std::atomic<int> s;
    std::vector<Nd4jLong> valuesX, valuesY;
    int iterations = 1000;
    s = 0;
    auto func = PRAGMA_THREADS_FOR {
        s++;
    };

    samediff::Threads::parallel_for(func, 0, 8192, 1, 4);

    ////////

    auto x = NDArrayFactory::create<float>('c', {32, 3, 256, 256});
    auto z = NDArrayFactory::create<float>('c', {32, 3, 256, 256});
    x.linspace(1.0);

    auto xs0 = x.sizeAt(0);
    auto xs1 = x.sizeAt(1);
    auto xs2 = x.sizeAt(2);
    auto xs3 = x.sizeAt(3);

    auto buffer = x.bufferAsT<float>();
    auto zbuffer = z.bufferAsT<float>();

    for (int e = 0; e < iterations; e++) {
        auto timeStart = std::chrono::system_clock::now();
        PRAGMA_OMP_PARALLEL_FOR_COLLAPSE(2)
        for (int i = 0; i < xs0; i++) {
            for (int j = 0; j < xs1; j++) {
                auto thread_id = omp_get_thread_num();
                for (int k = 0; k < xs2; k++) {
                    for (int l = 0; l < xs3; l++) {
                        zbuffer[thread_id] += buffer[i * j + (k*l)] * 2.5f;
                    }
                }
            }
        }
        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::nanoseconds>(timeEnd - timeStart).count();
        valuesX.emplace_back(outerTime);
    }


    for (int e = 0; e < iterations; e++) {
        auto timeStart = std::chrono::system_clock::now();
        auto f2d = PRAGMA_THREADS_FOR_2D {
            for (auto i = start_x; i < stop_x; i++) {
                for (auto j = start_y; j < stop_y; j++) {

                    for (auto k = 0; k < xs2; k++) {
                        for (auto l = 0; l < xs3; l++) {
                            zbuffer[thread_id] += buffer[i * j + (k * l)] * 2.5f;
                        }
                    }
                }
            }
        };
        samediff::Threads::parallel_for(f2d, 0, xs0, 1, 0, xs1, 1);

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::nanoseconds>(timeEnd - timeStart).count();
        valuesY.emplace_back(outerTime);
    }

    if (valuesX.size() > 0) {
        std::sort(valuesX.begin(), valuesX.end());
        nd4j_printf("OpenMP time: %lld; Min: %lld; Max: %lld;\n", valuesX[valuesX.size() / 2], valuesX[0], valuesX[valuesX.size() - 1]);
    }

    if (valuesY.size() > 0) {
        std::sort(valuesY.begin(), valuesY.end());
        nd4j_printf("Threads time: %lld; Min: %lld; Max: %lld;\n", valuesY[valuesY.size() / 2], valuesY[0], valuesY[valuesY.size() - 1]);
    }

    nd4j_printf("Sum: %f\n", z.sumNumber().e<float>(0));
}


TEST_F(PlaygroundTests, test_s_5) {
    auto x = NDArrayFactory::create<float>('c', {32, 1, 28, 28});

    std::vector<Nd4jLong> values;
    auto iterations = 100;

    auto startX = 0;
    auto stopX = x.sizeAt(0);
    auto incX = 1;
    auto startY = 0;
    auto stopY = x.sizeAt(1);
    auto incY = 1;
    auto numThreads = 4;

    // number of elements per loop
    auto delta_x = (stopX - startX);
    auto delta_y = (stopY - startY);

    // number of iterations per loop
    auto itersX = delta_x / incX;
    auto itersY = delta_y / incY;

    for (int e = 0; e < iterations; e++) {
        auto timeStart = std::chrono::system_clock::now();

        // picking best fit here
        auto splitLoop = samediff::ThreadsHelper::pickLoop2d(numThreads, itersX, itersY);
        auto span = samediff::Span2::build(splitLoop, 0, numThreads, startX, stopX, incX, startY, stopY, incY);

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::nanoseconds>(timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }

    std::sort(values.begin(), values.end());

    nd4j_printf("Calculations time: [Median: %lld; Min: %lld; Max: %lld;]\n", values[values.size() / 2], values[0], values[values.size()-1]);
}


TEST_F(PlaygroundTests, test_s_6) {
    auto x = NDArrayFactory::create<float>('c', {1024 * 1024 * 64});
    auto buffer = x.bufferAsT<float>();
    auto len = x.lengthOf();
    std::vector<Nd4jLong> values;
    auto iterations = 1000;

    for (int i = 0; i < iterations; i++) {
        auto timeStart = std::chrono::system_clock::now();

        // picking best fit here
        for (int e = 0; e < len; e++) {
            buffer[e] = (buffer[e] + 1.72f) * 3.17f - 0.0012f;
        }

        auto timeEnd = std::chrono::system_clock::now();
        auto outerTime = std::chrono::duration_cast<std::chrono::nanoseconds>(timeEnd - timeStart).count();
        values.emplace_back(outerTime);
    }

    std::sort(values.begin(), values.end());

    nd4j_printf("Calculations time: [Median: %lld; Min: %lld; Max: %lld;]\n", values[values.size() / 2], values[0], values[values.size()-1]);
}


TEST_F(PlaygroundTests, test_s_3) {
    std::atomic<int> s;
    s = 0;
    auto func = PRAGMA_THREADS_FOR {
        s++;
    };

    for (int e = 0; e < 10000; e++) {

        samediff::Threads::parallel_for(func, 0, 8192, 1, 4);
    }
}
 */

/*
TEST_F(PlaygroundTests, test_relubp_1) {
    auto x = NDArrayFactory::create<float>('c', {128, 64, 224, 224});
    auto y = x.ulike();
    auto z = x.ulike();
    RandomGenerator rng(119, 120);
    RandomLauncher::fillUniform(LaunchContext::defaultContext(), rng, &x, -1.0, 1.0);
    RandomLauncher::fillUniform(LaunchContext::defaultContext(), rng, &y, -1.0, 1.0);

    int iterations = 10;

    auto timeStart = std::chrono::system_clock::now();
    for (int e = 0; e < iterations; e++)
        ops::helpers::reluDerivative(LaunchContext::defaultContext(), &x, &y, &z);
    auto timeEnd = std::chrono::system_clock::now();

    auto outerTime = std::chrono::duration_cast<std::chrono::microseconds> (timeEnd - timeStart).count();
    auto time = (Nd4jLong) outerTime / iterations;
    auto bw = (1000000L * (float) (x.lengthOf() * x.sizeOfT()) / time) / 1024 / 1024 / 1024;

    nd4j_printf("Time: %lld; BW: %f GB/s\n", time, bw);
}

//////////////////////////////////////////////////////////////////////
TEST_F(PlaygroundTests, my) {

    int bS=8, iD=32,iH=32,iW=32,  iC=128,  kD=2,kH=2,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=2,dH=2,dW=2;
    int       oD,oH,oW;

    sd::ops::ConvolutionUtils::calcOutSizeDeconv3D(oD, oH, oW, kD, kH, kW, sD, sH, sW, pD, pH, pW, dD, dH, dW, iD, iH, iW, 0);

    printf("!!%i, %i, %i\n", oD,oH,oW);

    NDArray col('c', {bS, iC, kD, kH, kW, iD, iH, iW}, sd::DataType::DOUBLE);
    NDArray vol('c', {bS, iC, oD, oH, oW}, sd::DataType::DOUBLE);

    col = 3.77;
    vol = -10.33;

    auto variableSpace = new VariableSpace();
    auto block = new Context(1, variableSpace, false);  // not-in-place

    auto timeStart = std::chrono::system_clock::now();
    sd::ops::ConvolutionUtils::col2vol(*block, col, vol, sD, sH, sW, pD, pH, pW, dD, dH, dW);
    auto timeEnd = std::chrono::system_clock::now();
    auto time = std::chrono::duration_cast<std::chrono::microseconds> (timeEnd - timeStart).count();

    printf("time: %i \n", time);

    delete block;
    delete variableSpace;
}

TEST_F(PlaygroundTests, my) {

    int bS=32, iD=32,iH=64,iW=64,  iC=128,  kD=2,kH=2,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=2,dH=2,dW=2;
    int       oD,oH,oW;

    // sd::ops::ConvolutionUtils::calcOutSizeDeconv3D(oD, oH, oW, kD, kH, kW, sD, sH, sW, pD, pH, pW, dD, dH, dW, iD, iH, iW, 0);
    sd::ops::ConvolutionUtils::calcOutSizeDeconv2D(oH, oW, kH, kW, sH, sW, pH, pW,dH, dW, iH, iW, 0);

    printf("!!%i, %i, %i\n", oD,oH,oW);

    // NDArray col('c', {bS, iC, kD, kH, kW, iD, iH, iW}, sd::DataType::DOUBLE);
    // NDArray vol('c', {bS, iC, oD, oH, oW}, sd::DataType::DOUBLE);
    NDArray col('c', {bS, iC, kH, kW, iH, iW}, sd::DataType::DOUBLE);
    NDArray im('c', {bS, iC, oH, oW}, sd::DataType::DOUBLE);

    col = 3.77;
    // vol = -10.33;
    im = -10.33;

    auto variableSpace = new VariableSpace();
    auto block = new Context(1, variableSpace, false);  // not-in-place

    auto timeStart = std::chrono::system_clock::now();
    // sd::ops::ConvolutionUtils::col2vol(*block, col, vol, sD, sH, sW, pD, pH, pW, dD, dH, dW);
    sd::ops::helpers::col2im(*col.getContext(), col, im, sH, sW, pH, pW, iH, iW, dH, dW);
    auto timeEnd = std::chrono::system_clock::now();
    auto time = std::chrono::duration_cast<std::chrono::microseconds> (timeEnd - timeStart).count();

    printf("time: %i \n", time);

    delete block;
    delete variableSpace;
}

///////////////////////////////////////////////////////////////////
TEST_F(PlaygroundTests, lstmLayerCellBp_1) {

    const int bS   = 2;
    const int nIn  = 4;
    const int nOut = 3;
    // const int nIn  = 8;
    // const int nOut = 6;

    const float cellClip = 1.1;       // clipping value
    const Nd4jLong gateAct = 2;        // sigmoid activation for input (i), forget (f) and output (o) gates
    const float gateAlpha = 0;      // alpha value for activation for gates, not required for sigmoid
    const float gateBeta = 0;       // beta value for activation for gates, not required for sigmoid
    const Nd4jLong cellAct = 0;        // tanh activation for cell state
    const float cellAlpha = 0;      // alpha value for cell state activation, not required for tanh
    const float cellBeta = 0;       // beta value for cell state activation, not required for tanh
    const Nd4jLong outAct = 0;         // tanh activation for output
    const float outAlpha = 0;       // alpha value for output activation, not required for tanh
    const float outBeta = 0;        // beta value for output activation, not required for tanh

    NDArray x ('c',   {bS, nIn}, sd::DataType::DOUBLE);
    NDArray hI('c',   {bS, nOut}, sd::DataType::DOUBLE);
    NDArray cI('c',   {bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdh('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdc('c', {bS, nOut}, sd::DataType::DOUBLE);

    // NDArray x ('c',   {nIn}, sd::DataType::DOUBLE);
    // NDArray hI('c',   {nOut}, sd::DataType::DOUBLE);
    // NDArray cI('c',   {nOut}, sd::DataType::DOUBLE);
    // NDArray dLdh('c', {nOut}, sd::DataType::DOUBLE);
    // NDArray dLdc('c', {nOut}, sd::DataType::DOUBLE);

    NDArray Wx('c', {nIn, 4*nOut}, sd::DataType::DOUBLE);
    NDArray Wr('c', {nOut, 4*nOut}, sd::DataType::DOUBLE);
    NDArray b ('c', {4*nOut}, sd::DataType::DOUBLE);
    NDArray Wp('c', {3*nOut}, sd::DataType::DOUBLE);

    x.linspace(-4,1);
    hI.linspace(-2.5,0.5);
    cI.linspace(-3,0.5);
    Wx.linspace(0,0.1);
    Wr.linspace(3,-0.1);
    Wp.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    // x.assign(1.);
    // hI.assign(2.);
    // cI.assign(3.);
    // Wx.assign(0.5);
    // Wr.assign(0.5);
    // Wp.assign(0.75);
    // b.assign(0.7);

    std::vector<double>   tArgs = {cellClip};
    std::vector<Nd4jLong> iArgs = {gateAct, cellAct, outAct};

    // std::vector<bool>     bArgs = {false, false};
    // const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &hI, &cI}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &hI, &cI, &dLdh}, tArgs, iArgs, bArgs);

    std::vector<bool>     bArgs = {true, true};
    const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &b, &hI, &cI, &Wp}, tArgs, iArgs, bArgs);
    const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &hI, &cI, &Wp, &dLdh}, tArgs, iArgs, bArgs);

    sd::ops::lstmLayerCell opFF;
    sd::ops::lstmLayerCellBp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP, {true, true, true, true, true, true, true});
}

TEST_F(PlaygroundTests, my) {

    const int N = 40;

    NDArray x('c', {256,256,128,128}, sd::DataType::FLOAT32);
    NDArray z1('c', {256,2,128}, sd::DataType::DOUBLE);
    NDArray z = z1({0,0,0,1,0,0});
    z.printShapeInfo();

    auto timeStart = std::chrono::system_clock::now();
    for (int i = 0; i < N; ++i) {
        // x.reduceAlongDimension(sd::reduce::Mean, z, {1,3});
        x.applyBroadcast(sd::broadcast::Ops::Add, {1,3}, z, x);
    }
    auto timeEnd = std::chrono::system_clock::now();
    auto time = std::chrono::duration_cast<std::chrono::microseconds> ((timeEnd - timeStart) / N).count();

    printf("old %ld\n", time);
}


///////////////////////////////////////////////////////////////////
TEST_F(DeclarableOpsTests13, lstmLayer_bp_1) {

    const int sL   = 3;
    const int bS   = 2;
    const int nIn  = 2;
    const int nOut = 3;

    const int dataFormat = 0;       // [sL,bS,nIn]
    const int directionMode = 0;    // forward
    const int gateAct = 2;          // sigmoid activation for input (i), forget (f) and output (o) gates
    const int cellAct = 0;          // tanh activation for cell state
    const int outAct = 0;           // tanh activation for output

    const bool hasBiases  = true;   // biases array is provided
    const bool hasSeqLen  = false;  // seqLen array is not provided
    const auto hasInitH   = true;   // initial output is provided
    const auto hasInitC   = true;   // initial cell state is provided
    const auto hasPH      = true;   // peephole connections are absent
    const auto retFullSeq = true;   // dLdh per each time step
    const auto retLastH   = true;   // output at last time step
    const auto retLastC   = true;  // cells state at last time step

    const double cellClip = 0.5;       // clipping

    NDArray x('c', {sL, bS, nIn}, sd::DataType::DOUBLE);
    NDArray Wx('c', {nIn, 4*nOut}, sd::DataType::DOUBLE);
    NDArray Wr('c', {nOut, 4*nOut}, sd::DataType::DOUBLE);
    NDArray b('c', {4*nOut}, sd::DataType::DOUBLE);
    NDArray hI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray cI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray Wp('c', {3*nOut}, sd::DataType::DOUBLE);
    NDArray dLdh('c', {sL, bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdhL('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdcL('c', {bS, nOut}, sd::DataType::DOUBLE);

    x.linspace(-2,0.1);
    hI.linspace(-1.5,0.1);
    cI.linspace(0.7,-0.1);
    Wx.linspace(1,-0.1);
    Wr.linspace(-1,0.1);
    Wp.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    std::vector<double>   tArgs = {cellClip};
    std::vector<Nd4jLong> iArgs = {dataFormat, directionMode, gateAct, cellAct, outAct};
    std::vector<bool>     bArgs = {hasBiases, hasSeqLen, hasInitH, hasInitC, hasPH, retFullSeq, retLastH, retLastC};

    const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &b, &hI, &cI, &Wp}, tArgs, iArgs, bArgs);
    const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &hI, &cI, &Wp, &dLdh, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);

    sd::ops::lstmLayer opFF;
    sd::ops::lstmLayer_bp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP);

    ASSERT_TRUE(isGradCorrect);
}

///////////////////////////////////////////////////////////////////
TEST_F(DeclarableOpsTests13, lstmLayer_bp_2) {

    const int sL   = 3;
    const int bS   = 2;
    const int nIn  = 2;
    const int nOut = 3;

    const int dataFormat = 1;       // [bS,sL,nIn]
    const int directionMode = 0;    // forward
    const int gateAct = 2;          // sigmoid activation for input (i), forget (f) and output (o) gates
    const int cellAct = 0;          // tanh activation for cell state
    const int outAct = 0;           // tanh activation for output

    const bool hasBiases  = true;   // biases array is provided
    const bool hasSeqLen  = false;  // seqLen array is not provided
    const auto hasInitH   = true;   // initial output is provided
    const auto hasInitC   = true;   // initial cell state is provided
    const auto hasPH      = true;   // peephole connections are absent
    const auto retFullSeq = true;   // return whole h {h_0, h_1, ... , h_sL-1}, [sL,bS,nOut]
    const auto retLastH   = false;  // output at last time step
    const auto retLastC   = true;   // cells state at last time step

    const double cellClip = 0.5;       // clipping

    NDArray x('c', {bS, sL, nIn}, sd::DataType::DOUBLE);
    NDArray Wx('c', {nIn, 4*nOut}, sd::DataType::DOUBLE);
    NDArray Wr('c', {nOut, 4*nOut}, sd::DataType::DOUBLE);
    NDArray b('c', {4*nOut}, sd::DataType::DOUBLE);
    NDArray hI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray cI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray Wp('c', {3*nOut}, sd::DataType::DOUBLE);
    NDArray dLdh('c', {bS, sL, nOut}, sd::DataType::DOUBLE);
    NDArray dLdcL('c', {bS, nOut}, sd::DataType::DOUBLE);

    x.linspace(-2,0.1);
    hI.linspace(-1.5,0.1);
    cI.linspace(0.7,-0.1);
    Wx.linspace(1,-0.1);
    Wr.linspace(-1,0.1);
    Wp.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    std::vector<double>   tArgs = {cellClip};
    std::vector<Nd4jLong> iArgs = {dataFormat, directionMode, gateAct, cellAct, outAct};
    std::vector<bool>     bArgs = {hasBiases, hasSeqLen, hasInitH, hasInitC, hasPH, retFullSeq, retLastH, retLastC};

    const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &b, &hI, &cI, &Wp}, tArgs, iArgs, bArgs);
    const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &hI, &cI, &Wp, &dLdh, &dLdcL}, tArgs, iArgs, bArgs);

    sd::ops::lstmLayer opFF;
    sd::ops::lstmLayer_bp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP, std::vector<bool>(), {0., 1.}, GradCheck::LossFunc::MEAN);

    ASSERT_TRUE(isGradCorrect);
}

///////////////////////////////////////////////////////////////////
TEST_F(DeclarableOpsTests13, lstmLayer_bp_3) {

    const int sL   = 4;
    const int bS   = 3;
    const int nIn  = 3;
    const int nOut = 2;

    const int dataFormat = 2;       // [bS, nIn, sL]
    const int directionMode = 0;    // forward
    const int gateAct = 2;          // sigmoid activation for input (i), forget (f) and output (o) gates
    const int cellAct = 0;          // tanh activation for cell state
    const int outAct = 0;           // tanh activation for output

    const bool hasBiases  = true;   // biases array is provided
    const bool hasSeqLen  = true;  // seqLen array is not provided
    const auto hasInitH   = true;   // initial output is provided
    const auto hasInitC   = true;   // initial cell state is provided
    const auto hasPH      = true;   // peephole connections are absent
    const auto retFullSeq = true;   // dLdh per each time step
    const auto retLastH   = true;   // output at last time step
    const auto retLastC   = true;   // cells state at last time step

    const double cellClip = 0.5;       //  clipping

    NDArray x('c', {bS, nIn, sL}, sd::DataType::DOUBLE);
    NDArray Wx('c', {nIn, 4*nOut}, sd::DataType::DOUBLE);
    NDArray Wr('c', {nOut, 4*nOut}, sd::DataType::DOUBLE);
    NDArray b('c', {4*nOut}, sd::DataType::DOUBLE);
    NDArray seqLen('c', {bS}, {2,0,4}, sd::DataType::DOUBLE);
    NDArray hI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray cI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray Wp('c', {3*nOut}, sd::DataType::DOUBLE);
    NDArray dLdh('c', {bS, nOut, sL}, sd::DataType::DOUBLE);
    NDArray dLdhL('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdcL('c', {bS, nOut}, sd::DataType::DOUBLE);

    x.linspace(-2,0.1);
    hI.linspace(-1.5,0.1);
    cI.linspace(0.7,-0.1);
    Wx.linspace(1,-0.1);
    Wr.linspace(-1,0.1);
    Wp.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    std::vector<double>   tArgs = {cellClip};
    std::vector<Nd4jLong> iArgs = {dataFormat, directionMode, gateAct, cellAct, outAct};
    std::vector<bool>     bArgs = {hasBiases, hasSeqLen, hasInitH, hasInitC, hasPH, retFullSeq, retLastH, retLastC};

    const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp}, tArgs, iArgs, bArgs);
    const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);

    sd::ops::lstmLayer opFF;
    sd::ops::lstmLayer_bp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP, {true, true, true, true, false, true, true, true});

    ASSERT_TRUE(isGradCorrect);
}

///////////////////////////////////////////////////////////////////
TEST_F(DeclarableOpsTests13, lstmLayer_bp_4) {

    const int sL   = 3;
    const int bS   = 2;
    const int nIn  = 2;
    const int nOut = 3;

    const int dataFormat = 1;       // [bS,sL,nIn]
    const int directionMode = 1;    // backward
    const int gateAct = 2;          // sigmoid activation for input (i), forget (f) and output (o) gates
    const int cellAct = 0;          // tanh activation for cell state
    const int outAct = 0;           // tanh activation for output

    const bool hasBiases  = true;   // biases array is provided
    const bool hasSeqLen  = false;  // seqLen array is not provided
    const auto hasInitH   = true;   // initial output is provided
    const auto hasInitC   = true;   // initial cell state is provided
    const auto hasPH      = true;   // peephole connections are absent
    const auto retFullSeq = true;   // dLdh per each time step
    const auto retLastH   = true;   // output at last time step
    const auto retLastC   = true;   // cells state at last time step

    const double cellClip = 0.5;       // clipping

    NDArray x('c', {bS, sL, nIn}, sd::DataType::DOUBLE);
    NDArray Wx('c', {nIn, 4*nOut}, sd::DataType::DOUBLE);
    NDArray Wr('c', {nOut, 4*nOut}, sd::DataType::DOUBLE);
    NDArray b('c', {4*nOut}, sd::DataType::DOUBLE);
    NDArray hI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray cI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray Wp('c', {3*nOut}, sd::DataType::DOUBLE);
    NDArray dLdh('c', {bS, sL, nOut}, sd::DataType::DOUBLE);
    NDArray dLdhL('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdcL('c', {bS, nOut}, sd::DataType::DOUBLE);

    x.linspace(-2,0.1);
    hI.linspace(-1.5,0.1);
    cI.linspace(0.7,-0.1);
    Wx.linspace(1,-0.1);
    Wr.linspace(-1,0.1);
    Wp.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    std::vector<double>   tArgs = {cellClip};
    std::vector<Nd4jLong> iArgs = {dataFormat, directionMode, gateAct, cellAct, outAct};
    std::vector<bool>     bArgs = {hasBiases, hasSeqLen, hasInitH, hasInitC, hasPH, retFullSeq, retLastH, retLastC};

    const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &b, &hI, &cI, &Wp}, tArgs, iArgs, bArgs);
    const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &hI, &cI, &Wp, &dLdh, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);

    sd::ops::lstmLayer opFF;
    sd::ops::lstmLayer_bp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP);

    ASSERT_TRUE(isGradCorrect);
}

///////////////////////////////////////////////////////////////////
TEST_F(DeclarableOpsTests13, lstmLayer_bp_5) {

    const int sL   = 3;
    const int bS   = 2;
    const int nIn  = 2;
    const int nOut = 2;

    const int dataFormat = 2;       // [bS, nIn, sL]
    const int directionMode = 1;    // backward
    const int gateAct = 2;          // sigmoid activation for input (i), forget (f) and output (o) gates
    const int cellAct = 0;          // tanh activation for cell state
    const int outAct = 0;           // tanh activation for output

    const bool hasBiases  = true;   // biases array is provided
    const bool hasSeqLen  = true;  // seqLen array is not provided
    const auto hasInitH   = true;   // initial output is provided
    const auto hasInitC   = true;   // initial cell state is provided
    const auto hasPH      = true;   // peephole connections are absent
    const auto retFullSeq = true;   // dLdh per each time step
    const auto retLastH   = true;   // output at last time step
    const auto retLastC   = true;   // cells state at last time step

    const double cellClip = 0.5;       // clipping

    NDArray x('c', {bS, nIn, sL}, sd::DataType::DOUBLE);
    NDArray Wx('c', {nIn, 4*nOut}, sd::DataType::DOUBLE);
    NDArray Wr('c', {nOut, 4*nOut}, sd::DataType::DOUBLE);
    NDArray b('c', {4*nOut}, sd::DataType::DOUBLE);
    NDArray seqLen('c', {bS}, {0,2}, sd::DataType::DOUBLE);
    NDArray hI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray cI('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray Wp('c', {3*nOut}, sd::DataType::DOUBLE);
    NDArray dLdh('c', {bS, nOut, sL}, sd::DataType::DOUBLE);
    NDArray dLdhL('c', {bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdcL('c', {bS, nOut}, sd::DataType::DOUBLE);

    x.linspace(-2,0.1);
    hI.linspace(-1.5,0.1);
    cI.linspace(0.7,-0.1);
    Wx.linspace(1,-0.1);
    Wr.linspace(-1,0.1);
    Wp.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    std::vector<double>   tArgs = {cellClip};
    std::vector<Nd4jLong> iArgs = {dataFormat, directionMode, gateAct, cellAct, outAct};
    std::vector<bool>     bArgs = {hasBiases, hasSeqLen, hasInitH, hasInitC, hasPH, retFullSeq, retLastH, retLastC};

    const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp}, tArgs, iArgs, bArgs);
    const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);

    sd::ops::lstmLayer opFF;
    sd::ops::lstmLayer_bp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP, {true, true, true, true, false, true, true, true});

    ASSERT_TRUE(isGradCorrect);
}

///////////////////////////////////////////////////////////////////
TEST_F(DeclarableOpsTests13, lstmLayer_bp_6) {

    const int sL   = 3;
    const int bS   = 2;
    const int nIn  = 2;
    const int nOut = 2;

    const int dataFormat = 2;       // [bS, nIn, sL]
    const int directionMode = 2;    // bidirectional sum
    const int gateAct = 2;          // sigmoid activation for input (i), forget (f) and output (o) gates
    const int cellAct = 0;          // tanh activation for cell state
    const int outAct = 0;           // tanh activation for output

    const bool hasBiases  = true;   // biases array is provided
    const bool hasSeqLen  = true;  // seqLen array is not provided
    const auto hasInitH   = true;   // initial output is provided
    const auto hasInitC   = true;   // initial cell state is provided
    const auto hasPH      = true;   // peephole connections are absent
    const auto retFullSeq = true;   // dLdh per each time step
    const auto retLastH   = true;   // output at last time step
    const auto retLastC   = true;   // cells state at last time step

    const double cellClip = 0.5;       // clipping

    NDArray x('c', {bS, nIn, sL}, sd::DataType::DOUBLE);
    NDArray Wx('c', {2, nIn, 4*nOut}, sd::DataType::DOUBLE);
    NDArray Wr('c', {2, nOut, 4*nOut}, sd::DataType::DOUBLE);
    NDArray b('c', {2, 4*nOut}, sd::DataType::DOUBLE);
    NDArray seqLen('c', {bS}, {0,2}, sd::DataType::DOUBLE);
    NDArray hI('c', {2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray cI('c', {2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray Wp('c', {2, 3*nOut}, sd::DataType::DOUBLE);
    NDArray dLdh('c', {bS, nOut, sL}, sd::DataType::DOUBLE);
    NDArray dLdhL('c', {2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdcL('c', {2, bS, nOut}, sd::DataType::DOUBLE);

    x.linspace(-2,0.1);
    hI.linspace(-1.5,0.1);
    cI.linspace(0.7,-0.1);
    Wx.linspace(1,-0.1);
    Wr.linspace(-1,0.1);
    Wp.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    std::vector<double>   tArgs = {cellClip};
    std::vector<Nd4jLong> iArgs = {dataFormat, directionMode, gateAct, cellAct, outAct};
    std::vector<bool>     bArgs = {hasBiases, hasSeqLen, hasInitH, hasInitC, hasPH, retFullSeq, retLastH, retLastC};

    const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp}, tArgs, iArgs, bArgs);
    const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdhL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdcL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdhL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdcL}, tArgs, iArgs, bArgs);

    sd::ops::lstmLayer opFF;
    sd::ops::lstmLayer_bp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP, {true, true, true, true, false, true, true, true});

    ASSERT_TRUE(isGradCorrect);
}

///////////////////////////////////////////////////////////////////
TEST_F(DeclarableOpsTests13, lstmLayer_bp_7) {

    const int sL   = 3;
    const int bS   = 2;
    const int nIn  = 2;
    const int nOut = 2;

    const int dataFormat = 1;       // [bS,sL,nIn]
    const int directionMode = 3;    // bidirectional concat
    const int gateAct = 2;          // sigmoid activation for input (i), forget (f) and output (o) gates
    const int cellAct = 0;          // tanh activation for cell state
    const int outAct = 0;           // tanh activation for output

    const bool hasBiases  = true;   // biases array is provided
    const bool hasSeqLen  = true;  // seqLen array is not provided
    const auto hasInitH   = true;   // initial output is provided
    const auto hasInitC   = true;   // initial cell state is provided
    const auto hasPH      = true;   // peephole connections are absent
    const auto retFullSeq = true;   // dLdh per each time step
    const auto retLastH   = true;   // output at last time step
    const auto retLastC   = true;   // cells state at last time step

    const double cellClip = 0.5;       // clipping

    NDArray x('c', {bS,sL,nIn}, sd::DataType::DOUBLE);
    NDArray Wx('c', {2, nIn, 4*nOut}, sd::DataType::DOUBLE);
    NDArray Wr('c', {2, nOut, 4*nOut}, sd::DataType::DOUBLE);
    NDArray b('c', {2, 4*nOut}, sd::DataType::DOUBLE);
    NDArray seqLen('c', {bS}, {0,2}, sd::DataType::DOUBLE);
    NDArray hI('c', {2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray cI('c', {2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray Wp('c', {2, 3*nOut}, sd::DataType::DOUBLE);
    NDArray dLdh('c', {bS,sL,2*nOut}, sd::DataType::DOUBLE);
    NDArray dLdhL('c', {2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdcL('c', {2, bS, nOut}, sd::DataType::DOUBLE);

    x.linspace(-2,0.1);
    hI.linspace(-1.5,0.1);
    cI.linspace(0.7,-0.1);
    Wx.linspace(1,-0.1);
    Wr.linspace(-1,0.1);
    Wp.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    std::vector<double>   tArgs = {cellClip};
    std::vector<Nd4jLong> iArgs = {dataFormat, directionMode, gateAct, cellAct, outAct};
    std::vector<bool>     bArgs = {hasBiases, hasSeqLen, hasInitH, hasInitC, hasPH, retFullSeq, retLastH, retLastC};

    const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp}, tArgs, iArgs, bArgs);
    const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdhL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdcL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdhL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdcL}, tArgs, iArgs, bArgs);

    sd::ops::lstmLayer opFF;
    sd::ops::lstmLayer_bp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP, {true, true, true, true, false, true, true, true});

    ASSERT_TRUE(isGradCorrect);
}

///////////////////////////////////////////////////////////////////
TEST_F(DeclarableOpsTests13, lstmLayer_bp_8) {

    const int sL   = 3;
    const int bS   = 2;
    const int nIn  = 2;
    const int nOut = 2;

    const int dataFormat = 3;       // [sL, bS, nIn]
    const int directionMode = 4;    // bidirectional extra output dim
    const int gateAct = 2;          // sigmoid activation for input (i), forget (f) and output (o) gates
    const int cellAct = 0;          // tanh activation for cell state
    const int outAct = 0;           // tanh activation for output

    const bool hasBiases  = true;   // biases array is provided
    const bool hasSeqLen  = true;  // seqLen array is not provided
    const auto hasInitH   = true;   // initial output is provided
    const auto hasInitC   = true;   // initial cell state is provided
    const auto hasPH      = true;   // peephole connections are absent
    const auto retFullSeq = true;   // dLdh per each time step
    const auto retLastH   = true;   // output at last time step
    const auto retLastC   = true;   // cells state at last time step

    const double cellClip = 0.5;       // clipping

    NDArray x('c', {sL, bS, nIn}, sd::DataType::DOUBLE);
    NDArray Wx('c', {2, nIn, 4*nOut}, sd::DataType::DOUBLE);
    NDArray Wr('c', {2, nOut, 4*nOut}, sd::DataType::DOUBLE);
    NDArray b('c', {2, 4*nOut}, sd::DataType::DOUBLE);
    NDArray seqLen('c', {bS}, {0,2}, sd::DataType::DOUBLE);
    NDArray hI('c', {2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray cI('c', {2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray Wp('c', {2, 3*nOut}, sd::DataType::DOUBLE);
    NDArray dLdh('c', {sL, 2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdhL('c', {2, bS, nOut}, sd::DataType::DOUBLE);
    NDArray dLdcL('c', {2, bS, nOut}, sd::DataType::DOUBLE);

    x.linspace(-2,0.1);
    hI.linspace(-1.5,0.1);
    cI.linspace(0.7,-0.1);
    Wx.linspace(1,-0.1);
    Wr.linspace(-1,0.1);
    Wp.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    std::vector<double>   tArgs = {cellClip};
    std::vector<Nd4jLong> iArgs = {dataFormat, directionMode, gateAct, cellAct, outAct};
    std::vector<bool>     bArgs = {hasBiases, hasSeqLen, hasInitH, hasInitC, hasPH, retFullSeq, retLastH, retLastC};

    const OpArgsHolder argsHolderFF({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp}, tArgs, iArgs, bArgs);
    const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdhL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh, &dLdcL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdhL, &dLdcL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdh}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdhL}, tArgs, iArgs, bArgs);
    // const OpArgsHolder argsHolderBP({&x, &Wx, &Wr, &b, &seqLen, &hI, &cI, &Wp, &dLdcL}, tArgs, iArgs, bArgs);

    sd::ops::lstmLayer opFF;
    sd::ops::lstmLayer_bp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP, {true, true, true, true, false, true, true, true});

    ASSERT_TRUE(isGradCorrect);
}

//////////////////////////////////////////////////////////////////////
TEST_F(DeclarableOpsTests15, gru_bp_1) {

    const int sL = 3;
    const int bS = 2;
    const int nIn = 5;
    const int nOut = 4;


    NDArray x('c', {sL, bS, nIn}, {0.5,  1. ,  1.5,  2. ,  2.5, 3. ,  3.5,  4. ,  4.5,  5. ,  5.5,  6. ,  6.5,  7. ,  7.5, 8. ,  8.5,  9. ,  9.5, 10. ,  10.5, 11. , 11.5, 12. , 12.5, 13. , 13.5, 14. , 14.5, 15.}, sd::DataType::DOUBLE);
    NDArray hI('c', {bS, nOut}, {-3,-2,-1,0,1,2,3,4}, sd::DataType::DOUBLE);
    NDArray Wx('c', {nIn, 3*nOut}, sd::DataType::DOUBLE);
    NDArray Wh('c', {nOut, 3*nOut}, sd::DataType::DOUBLE);
    NDArray b('c', {3*nOut}, sd::DataType::DOUBLE);

    NDArray dLdh('c', {sL, bS, nOut}, sd::DataType::DOUBLE);

    Wx.linspace(1,-0.1);
    Wh.linspace(0.2,0.2);
    b.linspace(1,-0.15);

    const OpArgsHolder argsHolderFF({&x, &hI, &Wx, &Wh, &b}, {}, {});
    const OpArgsHolder argsHolderBP({&x, &hI, &Wx, &Wh, &b, &dLdh}, {}, {});

    sd::ops::gru opFF;
    sd::ops::gru_bp opBP;

    const bool isGradCorrect = GradCheck::checkGrad(opFF, opBP, argsHolderFF, argsHolderBP);
}

*/