cavis/libnd4j/tests_cpu/layers_tests/ConvolutionTests1.cpp

/*******************************************************************************
 * Copyright (c) 2015-2018 Skymind, Inc.
 *
 * 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
 ******************************************************************************/

//
// @author raver119@gmail.com
//

#ifndef LIBND4J_CONVOLUTIONTESTS1_H
#define LIBND4J_CONVOLUTIONTESTS1_H

#include "testlayers.h"
#include <NDArray.h>
#include <Context.h>
#include <Node.h>
#include <graph/Variable.h>
#include <graph/VariableSpace.h>
#include <ops/declarable/CustomOperations.h>
#include <ops/declarable/helpers/convolutions.h>
#include <ops/declarable/helpers/col2im.h>
#include <PointersManager.h>
#include <GradCheck.h>

#ifdef HAVE_MKLDNN
#include <ops/declarable/platform/mkldnn/mkldnnUtils.h>
#endif

using namespace nd4j;
using namespace nd4j::graph;

class ConvolutionTests1 : public testing::Test {
public:

};

template <typename T>
class TypedConvolutionTests1 : public testing::Test {
public:

};

typedef ::testing::Types<double, float> TestingTypes;
TYPED_TEST_CASE(TypedConvolutionTests1, TestingTypes);

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_1) {

    int bS=1, iH=5,iW=4,  iC=2,oC=3,  kH=2,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    TypeParam _expB[]{664.0, 700.0, 736.0, 344.0, 808.0, 844.0, 880.0, 408.0, 952.0, 988.0, 1024.0, 472.0, 1096.0, 1132.0, 1168.0, 536.0, 466.0, 480.0, 494.0, 220.0, 1528.0, 1628.0, 1728.0, 856.0, 1928.0, 2028.0, 2128.0, 1048.0, 2328.0, 2428.0, 2528.0, 1240.0, 2728.0, 2828.0, 2928.0, 1432.0, 1346.0, 1392.0, 1438.0, 700.0, 2392.0, 2556.0, 2720.0, 1368.0, 3048.0, 3212.0, 3376.0, 1688.0, 3704.0, 3868.0, 4032.0, 2008.0, 4360.0, 4524.0, 4688.0, 2328.0, 2226.0, 2304.0, 2382.0, 1180.0};
    Nd4jLong _expS[]{4, 1, 3, 5, 4, 60, 20, 4, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99};
    auto input = NDArrayFactory::create_<TypeParam>('c', {bS, iC, iH, iW});
    auto weights = NDArrayFactory::create_<TypeParam>('c', {oC, iC, kH, kW});
    for (int e = 0; e < input->lengthOf(); e++)
        input->p(e, e + 1);

    for (int e = 0; e < weights->lengthOf(); e++)
        weights->p(e, e + 1);
    weights->permutei({2,3,1,0});

    // weights->printShapeInfo("weights");

    ArrayOptions::setDataType(_expS, input->dataType());
    auto exp = new NDArray(_expB, _expS);

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, input);
    variableSpace->putVariable(-2, weights);

    auto block = new Context(1, variableSpace, false);  // not-in-place
    block->fillInputs({-1, -2});
    // 5,5 kernel
    block->getIArguments()->push_back(kH);
    block->getIArguments()->push_back(kW);

    // 1,1 stride
    block->getIArguments()->push_back(sH);
    block->getIArguments()->push_back(sW);

    // 0,0 padding
    block->getIArguments()->push_back(pH);
    block->getIArguments()->push_back(pW);

    // 1,1 dilation
    block->getIArguments()->push_back(dH);
    block->getIArguments()->push_back(dW);

    // same mode
    block->getIArguments()->push_back(1);

    // is NHWC
    block->getIArguments()->push_back(0);

    nd4j::ops::conv2d op;

    Nd4jStatus status = op.execute(block);
    ASSERT_EQ(ND4J_STATUS_OK, status);

    auto res = variableSpace->getVariable(1)->getNDArray();


    // checking output shape
    ASSERT_EQ(1, res->sizeAt(0));
    ASSERT_EQ(3, res->sizeAt(1));
    ASSERT_EQ(5, res->sizeAt(2));
    ASSERT_EQ(4, res->sizeAt(3));

    // basically the same as above
    ASSERT_TRUE(res->isSameShape(exp));
    // just for visual validation
    // exp->printIndexedBuffer("Expected");
    // res->printIndexedBuffer("Actual  ");
    // res->printShapeInfo("Result shape");
    // final check
    ASSERT_TRUE(res->equalsTo(exp));

    delete block;
    delete variableSpace;
    delete exp;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_2) {
    auto input = NDArrayFactory::create<TypeParam>('c', {1, 1, 1, 4});
    auto weights = NDArrayFactory::create<TypeParam>('c', {1, 1, 1, 4});
    auto exp = NDArrayFactory::create<TypeParam>('c', {1, 4, 1, 4}, {2.f, 4.f, 6.f, 8.f, 2.f, 4.f, 6.f, 8.f, 2.f, 4.f, 6.f, 8.f, 2.f, 4.f, 6.f, 8.f});

    weights.assign(2.0);
    input.linspace(1);

    nd4j::ops::conv2d op;
    auto result = op.execute({&input, &weights}, {}, {1, 1, 1, 1, 0, 0, 1, 1, 0, 0});
    ASSERT_EQ(ND4J_STATUS_OK, result->status());

    auto z = result->at(0);

    ASSERT_TRUE(exp.isSameShape(z));
    ASSERT_TRUE(exp.equalsTo(z));

    delete result;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_3) {

    int bS=2, iH=4,iW=3,  iC=4,oC=3,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=4,oW=3;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto weights = NDArrayFactory::create<TypeParam>('c', {kH, kW, iC, oC});
    auto bias = NDArrayFactory::create<TypeParam>('c', {oC}, {1.f, 2.f, 3.f});


    auto expOutput = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, oC},{ 152.f,  155.2f,  158.4f, 152.f,  155.2f,  158.4f,  66.4f,   68.f,   69.6f, 170.4f,  175.2f,  180.f, 170.4f,  175.2f,  180.f,  70.8f,   73.2f,   75.6f,
                                                      170.4f,  175.2f,  180.f, 170.4f,  175.2f,  180.f,  70.8f,   73.2f,   75.6f,  75.2f,   78.4f,   81.6f,  75.2f,   78.4f,   81.6f,  28.f,   29.6f,   31.2f,
                                                      152.f,  155.2f,  158.4f, 152.f,  155.2f,  158.4f,  66.4f,   68.f,   69.6f, 170.4f,  175.2f,  180.f, 170.4f,  175.2f,  180.f,  70.8f,   73.2f,   75.6f,
                                                      170.4f,  175.2f,  180.f, 170.4f,  175.2f,  180.f,  70.8f,   73.2f,   75.6f,  75.2f,   78.4f,   81.6f,  75.2f,   78.4f,   81.6f,  28.f,   29.6f,   31.2f});
    input = 2.;
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_4) {

    int bS=2, iH=4,iW=3,  iC=4,oC=3,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto weights = NDArrayFactory::create<TypeParam>('c', {kH, kW, iC, oC});
    auto bias = NDArrayFactory::create<TypeParam>('c', {oC}, {1,2,3});

    auto expOutput = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, oC},{ 170.4f,175.20001f,180.f,170.4f,175.20001f,180.f,170.4f,175.20001f,180.f,170.4f,175.20001f,180.f,170.4f,175.20001f,180.f,170.4f,175.20001f,180.f,170.4f,175.20001f,180.f,170.4f,175.20001f,180.f});

    input = 2.;
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_5) {

    int bS=2, iH=4,iW=3,  iC=4,oC=3,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {oC, iC, kH, kW});
    auto bias     = NDArrayFactory::create<TypeParam>('c', {oC}, {1,2,3});

    auto expOutput = NDArrayFactory::create<TypeParam>('c', {bS, oC, oH, oW}, {61.f,   61.f,  61.f,   61.f, 177.2f,  177.2f, 177.2f,  177.2f, 293.4f,  293.4f, 293.4f,  293.4f,  61.f,   61.f,  61.f,   61.f, 177.2f,  177.2f, 177.2f,  177.2f, 293.4f,  293.4f, 293.4f,  293.4f});

    input = 2.;
    weights.linspace(0.1, 0.1);
    weights.permutei({2,3,1,0});

    nd4j::ops::conv2d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto output = results->at(0);

    // output->printIndexedBuffer();

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_6) {
    auto input = NDArrayFactory::create<TypeParam>('c', {54, 1, 12, 12});
    auto weights = NDArrayFactory::create<TypeParam>('c', {1, 2, 12, 2});

    nd4j::ops::conv2d op;
    auto result = op.execute({&input, &weights}, {}, {-1,-1,  1,1,  0,0,  1,1,  1,1});
    ASSERT_EQ(Status::OK(), result->status());

    delete result;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_7) {

    int bS=1, iH=256,iW=256,  iC=1,oC=1,  kH=4,kW=3,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    // int       oH=256,oW=256;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kH, kW, iC, oC});

    input = 5.;
    weights = 3.;

    nd4j::ops::conv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, conv2d_8) {

    int bS=1, iH=6,iW=8,  iC=2,oC=2,  kH=2,kW=1,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=6,oW=8;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iC, iH, iW}, {0.679350, 0.355087, 0.842789, 0.200313, 0.701499, 0.310693, 0.447940, 0.938010, 0.326674, 0.151873, 0.383318, 0.782123, 0.198807,
        0.798564, 0.163263, 0.146968, 0.260897, 0.135058, 0.756209, 0.275454, 0.369088, 0.092826, 0.836492, 0.268413, 0.095062, 0.312795, 0.135918, 0.517544, 0.328703,
        0.061736, 0.396431, 0.248016, 0.548959, 0.115046, 0.814362, 0.721564, 0.404494, 0.299089, 0.403884, 0.988311, 0.022296, 0.927782, 0.318416, 0.068546, 0.284533,
        0.232720, 0.352142, 0.058909, 0.711221, 0.674457, 0.196946, 0.699497, 0.074322, 0.420425, 0.584263, 0.149574, 0.446406, 0.723072, 0.064481, 0.483078, 0.875996,
        0.569819, 0.445863, 0.527755, 0.016646, 0.753678, 0.140636, 0.754129, 0.161932, 0.775037, 0.332645, 0.117394, 0.017711, 0.608476, 0.525152, 0.917194, 0.849891,
        0.589423, 0.852278, 0.390636, 0.889683, 0.669445, 0.698873, 0.961480, 0.157401, 0.157364, 0.493520, 0.569937, 0.126832, 0.115728, 0.786368, 0.737939, 0.490079, 0.608414, 0.956500, 0.390098});

    NDArray weights('c', {kH, kW, iC, oC}, {0.07581716775894165, 0.8706002235412598, 0.29345420002937317, 0.5281786322593689, 0.10540834069252014, 0.3663792014122009, 0.17209206521511078, 0.6257694959640503});
    NDArray bias('c', {1, oC}, {0.7414038777351379, 0.8980839848518372});

    NDArray expOutput('c', {bS, oC, oH, oW}, {1.112878, 1.106691, 0.914598, 1.127438, 0.988108, 1.070572, 1.040759, 0.962728, 0.927537, 1.109045, 0.893301, 1.101278, 1.080314,
        1.112327, 1.030041, 0.955914, 0.779137, 1.110499, 0.944709, 1.195986, 0.997814, 1.083822, 1.090898, 0.889572, 0.964781, 1.071012, 1.111928, 1.291319, 1.085454, 0.977661,
        1.149068, 1.077099, 1.068283, 1.064290, 1.177125, 1.212480, 0.932593, 0.939493, 1.118576, 1.056927, 0.780314, 0.845707, 0.996308, 0.963152, 0.906792, 0.937590, 1.048791,
        0.860346, 2.264212, 2.071576, 1.916629, 2.030785, 2.169075, 2.039786, 1.935480, 2.177816, 1.524273, 1.933327, 1.630923, 2.406983, 1.770406, 2.413284, 1.790349, 1.476586,
        1.179925, 1.909109, 2.009143, 2.299778, 1.957207, 1.779718, 2.480604, 1.529086, 1.748063, 1.952856, 2.029487, 2.699131, 1.879842, 1.471205, 2.150177, 2.039078, 1.933456,
        1.764169, 2.584944, 2.521004, 1.744296, 1.707578, 2.237938, 2.325231, 0.984485, 1.766936, 1.590640, 1.347524, 1.404648, 1.422042, 1.709862, 1.155412});

    nd4j::ops::conv2d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto output = results->at(0);

    // output->printIndexedBuffer();

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

TYPED_TEST(TypedConvolutionTests1, TestAvgFF_TF) {

    auto input = NDArrayFactory::create<TypeParam>('c', {4, 10, 10, 3}, {9.37125111f, 2.20166993f, 2.91434479f, 5.43639755f, -2.10573769f, 4.08528662f, 5.86908436f, -4.46203756f, 2.21057916f, 5.35849190f, 0.01394637f, 4.40566349f, 7.07982206f, -0.09633455f, 2.42429352f, 3.97301817f, -1.89553940f, 1.99690318f, 6.33141708f, 0.55401880f, 1.70707977f,
        5.55204201f, -0.03513752f, 1.60011971f, 2.62700319f, -2.74582434f, 3.06697464f, 1.06277943f, -1.16075921f, -0.78095782f, 9.72352791f, -1.22686064f, 1.99644792f, 7.35571337f, 1.40607321f, 0.11390255f, 9.53334427f, 2.28303599f, -1.66728830f, 6.16678810f, -0.04532295f, -1.97708666f, 9.74906158f, 1.46223176f, -1.46734393f, 4.30761862f,
        -1.23790228f, 1.24823606f, 6.13938427f, -3.83689475f, -1.19625473f, 7.91535568f, 6.05868721f, -3.22946382f, 8.81633949f, -0.19967777f, 0.66053957f, 2.30919123f, 0.74543846f, -0.39347672f, 11.11058044f, 0.53720862f, 1.52645731f, 5.70012379f, -1.15213466f, 1.16451406f, 7.00526333f, 1.57362783f, -2.44384766f, 5.54213285f, -1.98828590f,
        -0.70483637f, 7.88281822f, -3.59875536f, 0.80745387f, 13.41578484f, -1.55507684f, -0.65855008f, 9.32583523f, -0.14544789f, 0.73436141f, 3.61176538f, -1.71268058f, -2.58490300f, 9.09280205f, -3.27405524f, -2.04569697f, 4.44761324f, -0.62955856f, -2.61917663f, 8.04890442f, 0.54579324f, 0.85929775f, 9.82259560f, -1.93825579f, 0.77703512f,
        4.67090321f, -4.79267597f, -2.38906908f, 9.31265545f, 0.96026313f, -1.14109385f, 11.54231834f, -0.01417295f, -0.39500344f, 8.49191666f, 0.55300158f, 2.79490185f, 6.92466164f, 1.72254205f, 2.82222271f, 8.83112717f, 2.95033407f, 2.18054962f, 6.73509789f, -2.22272944f, 0.51127720f, -1.04563558f, 2.15747333f, -2.30959272f, 9.55441570f,
        1.50396204f, 1.77370787f, 7.38146257f, -1.79076433f, 3.20961165f, 7.18864202f, 2.91217351f, 0.43018937f, 7.11078024f, -1.17386127f, -0.16817921f, 6.12327290f, -2.82205725f, 3.30696845f, 13.51291752f, -1.30856836f, -2.38332748f, 11.09487438f, -1.47190213f, -0.53050828f, 4.38285351f, -5.07309771f, 1.50714362f, 5.72274446f, -2.85825086f,
        -0.89673209f, 3.73791552f, -0.67708802f, -4.13149452f, -0.00671843f, -0.26566532f, 0.32961160f, 7.14501762f, -1.41608179f, -4.96590328f, 12.26205540f, -0.65158135f, -0.88641000f, 6.95777559f, -0.79058206f, -0.10260171f, 7.87169170f, 1.35921454f, 1.11759663f, 5.46187401f, -2.57214499f, 2.48484039f, 4.04043484f, -2.07137156f, -1.42709637f,
        9.25487137f, -0.12605135f, -2.66949964f, 2.89412403f, 0.74451172f, -2.96250391f, 3.99258423f, 0.27084303f, 0.32213116f, 5.42332172f, -0.44414216f, 1.70881832f, 6.69346905f, 0.53058422f, -4.73146200f, 4.22051668f, 2.24834967f, 0.66996074f, 4.30173683f, 0.11849818f, -4.07520294f, 8.27318478f, -2.54398274f, -2.86705542f, 10.11775303f,
        -0.99382895f, 0.65881538f, 7.93556786f, -1.27934420f, -1.69343162f, 9.68042564f, -1.02609646f, -1.18189347f, 5.75370646f, -1.67888868f, -4.48871994f, 4.79537392f, -0.79212248f, -0.19855022f, 6.15060997f, -0.01081491f, 3.64454579f, 10.82562447f, 1.58859253f, -2.65847278f, 8.60093212f, -1.59196103f, 0.07635692f, 11.76175690f, -1.17453325f,
        0.10122013f, 6.86458445f, -2.18891335f, -2.74004745f, 8.07066154f, 0.71818852f, -2.03035975f, 6.31053686f, 0.51509416f, 1.39789927f, 9.43515587f, 2.04256630f, 0.13985133f, 4.65010691f, 2.40911126f, -0.36255789f, -3.06867862f, -0.45225358f, -1.56778407f, 6.05917358f, -1.09891272f, 1.77184200f, 6.46248102f, 0.96042323f, -0.24346280f,
        4.63436460f, -4.69907761f, 1.25187206f, 11.46173859f, -2.21917558f, 1.28007793f, 6.92173195f, 2.11268163f, -3.47389889f, 5.08722782f, -3.03950930f, -4.17154264f, 11.30568314f, 0.80361372f, 2.53214502f, 7.18707085f, -4.49114513f, 2.85449266f, 10.14906883f, -0.31974933f, -0.84472644f, -0.52459574f, 0.12921631f, -1.81390119f, 2.76170087f, 1.03982210f, 2.91744232f, -0.29048753f, 5.87453508f, -1.53684759f, 1.85800636f, -0.91404629f, 1.28954852f, 5.11354685f, -2.47475505f, -1.33179152f, 2.58552408f, 1.37316465f, -3.32339454f, 1.54122913f, 3.24953628f, -0.29758382f, 2.82391763f, -1.51142192f, -1.22699404f, 6.75745535f, 0.65452754f, -3.29385471f, 2.06008053f, 2.53172946f, -4.23532820f, -1.53909743f, -0.07010663f, -1.42173731f, 7.29031610f, -0.18448229f, 4.59496164f, 6.73027277f, 0.73441899f, 0.14426160f, 4.14915276f, -2.97010231f, 6.05851364f, 4.95218086f, -2.39145470f, 2.40494704f, 2.10288811f, 0.53503096f, 1.44511235f, 6.66344261f, -3.05803776f, 7.21418667f, 3.30303526f, -0.24163735f, 3.47409391f, 3.64520788f, 2.15189481f, -3.11243272f, 3.62310791f, 0.37379482f, 0.40865007f, -0.83132005f, -4.78246069f, 2.07030797f, 6.51765442f, 3.16178989f, 5.06180477f, 3.78434467f, -0.96689719f, 0.35965276f, 5.89967585f, 1.40294051f, 1.11952639f, 10.59778214f, 0.26739889f, -1.61297631f, 6.24801159f, -0.93914318f, -0.57812452f, 9.92604542f, -0.73025000f, -3.38530874f, 2.45646000f, -2.47949195f, 0.51638460f, 10.65636063f, 1.97816694f, -3.00407791f, 2.66914415f, -0.81951088f, -0.23316640f, 2.40737987f, -2.70007610f, 1.51531935f, 4.08860207f, -0.27552786f, -1.31721711f, 7.11568260f, -3.33498216f, -4.02545023f, 7.22675610f, -0.81690705f, -2.52689576f, 1.04016697f, -0.79291463f, -0.34875512f, 10.00498390f, -4.24167728f, 1.46162593f, 11.82569408f, -1.70359993f, -0.30161047f, 16.44085884f, -0.82253462f, -0.09435523f, 6.13080597f, -0.20259480f, 0.68308711f, 6.15663004f, -6.61776876f, 0.33295766f, 2.55449438f, -0.17819691f, -1.14892209f, 5.56776142f, 1.99279118f, 1.33035934f, 4.45823956f, 3.34916544f, -2.59905386f, 6.16164446f, -2.03881931f, -2.45273542f, 12.46793365f, -2.22743297f, 2.83738565f, 8.48628139f, -1.39347959f, -1.30867767f, 11.08041477f, -4.00363779f, 2.09183025f, 11.30395889f, -2.20504737f, 1.37426853f, 8.98735619f, 1.04676604f, -0.72757077f, 8.28050232f, -6.70741081f, -0.65798020f, 5.68592072f, -0.60760021f, 0.35854483f, 6.26852131f, 1.94100165f, 1.32112014f, 0.80987954f, -1.74617672f, -0.25434083f, 7.16045523f, 1.58884013f, -2.64847064f, 13.14820385f, 1.21393633f, -2.47258949f, 9.41650105f, -0.79384226f, 2.48954105f, 10.95629311f, 0.47723705f, 4.02126694f, 8.02593136f, -2.20726371f, -1.18794477f, 1.50836647f, 0.93118095f, -1.73513174f, 8.85493565f, -2.99670315f, -0.79055870f, 2.39473820f, 2.05046916f, -2.38055134f, 11.82299423f, 0.15609655f, 0.68744308f, 5.66401434f, -0.69281673f, 2.09855556f, 7.74626589f, -0.34283102f, 1.00542057f, 9.95838642f, 0.80161905f, 2.33455157f, 9.80057335f, -0.93561798f, 2.56991577f, 8.29711342f, 0.94213426f, 0.44209945f, 11.70259857f, 0.92710167f, 2.60957146f, 0.24971688f, -0.86529571f, 3.78628922f, 6.80884457f, -0.68178189f, 2.21103406f, 3.18895817f, 0.60283208f, -2.92716241f, 6.72060776f, -1.06625068f, 2.56543374f, 9.97404480f, 3.58080721f, -0.94936347f, 10.16736984f, -1.38464379f, 1.18191063f, 6.66179037f, -3.56115270f, 0.32329530f, 10.90870762f, 2.20638227f, 0.19653285f, 7.34650040f, -3.63859272f, -1.03027737f, 5.98829985f, -3.66606474f, -3.89746714f, 8.63469028f, 1.22569811f, 1.63240814f, 3.74385309f, 0.58243257f, -0.56981975f, 3.69260955f, 1.00979900f, -1.44030499f, 8.57058144f, -1.10648811f, 1.20474911f, 5.43133020f, -2.14822555f, -0.07928789f, 11.25825310f, 0.19645604f, -5.49546146f, 10.41917038f, -0.68178523f, -2.99639869f, 6.50054455f, 0.46488351f, -5.42328453f, 9.09500027f, -2.82107449f, 0.05601966f, 15.34610748f, -0.06820253f, 3.86699796f, 10.73316956f, -3.04795432f, -0.14702171f, 5.64813185f, 1.44028485f, -2.47596145f, 0.07280898f, -3.03187990f, -1.35183525f, 9.35835648f, 2.72966957f, 1.88199532f, 10.36187744f, -0.22834805f, -3.26738238f, 6.92025137f, -2.34061313f, 4.77379704f, 5.28559113
    auto exp = NDArrayFactory::create<TypeParam>('c', {4, 4, 4, 3}, {7.97172260f,  0.06878620f,              2.27749538f,              7.29276514f,              -0.14074677f,              0.65480286f,              5.70313978f,              -0.06546132f,              0.35443667f,              3.70382833f,              -0.84020567f,              0.63826996f,              8.60301399f,              -0.38236514f,              1.55177069f,              7.37542057f,              -0.99374938f,              -0.29971302f,              8.84352493f,              -0.67121059f,              0.43132120f,              4.78175592f,              -1.25070143f,              -1.91523600f,              6.03855371f,              -0.00292124f,              -1.11214364f,              7.90158176f,              -0.57949901f,              -0.96735370f,              7.81192017f,              -0.53255427f,              -0.48009714f,              3.16953635f,              0.08353355f,              -1.54299748f,              3.74821687f,              1.69396687f,              0.72724354f,              5.42915201f,              -1.13686812f,              -0.71793109f,              5.78376389f,              -0.72239977f,              -0.60055625f,              2.53636408f,              0.56777251f,              -2.07892323f,              6.08064651f,              0.68620735f,              2.54017019f,              5.65828180f,              -0.68255502f,              1.47283304f,              6.10842514f,              -0.39655915f,              0.28380761f,              1.96707797f,              -1.98206317f,              0.94027776f,              4.71811438f,              0.32104525f,              -0.92409706f,              8.34588146f,              -1.05581069f,              -0.55217457f,              9.58440876f,              -0.96549922f,              0.45820439f,              5.65453672f,              -2.50953507f,              -0.71441835f,              8.03059578f,              -0.21281289f,              0.92125505f,              9.26900673f,              -0.35963219f,              -0.70039093f,              8.59924412f,              -1.22358346f,              0.81318003f,              3.85920119f,              -0.01305223f,              -1.09234154f,              6.33158875f,              1.28094780f,              -1.48926139f,              4.94969177f,              -0.77126902f,              -1.97033751f,              5.64381838f,              -0.16285487f,              -1.31277227f,              2.39893222f,              -1.32902908f,              -1.39609122f,              6.47572327f,              -0.45267010f,              1.55727172f,              6.70965624f,              -1.68735468f,              -0.05672536f,              7.25092363f,              -0.64613032f,              0.67050058f,              3.60789680f,              -2.05948973f,              2.22687531f,              8.15202713f,              -0.70148355f,              1.28314006f,              8.14842319f,              -1.88807654f,              -1.04808438f,              8.45500565f,              -0.76425624f,              0.94542569f,              4.56179953f,              -0.28786001f,              -2.04502511f,              8.46278095f,              -0.31019822f,              0.07339200f,              9.34214592f,              -0.61948007f,              0.52481830f,              8.32515621f,              -1.52418160f,              0.49678251f,              5.11082315f,              -1.09908783f,              -0.52969611f,              5.27806664f,              0.88632923f,              0.66754371f,              4.75839233f,              0.48928693f,              -0.68036932f,              6.56925392f,              -0.02949905f,              -2.99189186f,              4.46320581f,              -0.64534980f,              -0.29516968f,              8.60809517f,              -1.13120568f,              3.41720533f,              5.84243155f,              -1.24109328f,              0.89566326f,              5.99578333f,              -0.42496428f,              2.07076764f,       

    nd4j::ops::avgpool2d op;
    auto result = op.execute({&input}, {}, {3,3,  3,3,  0,0,  1,1,1,  0,1}, {});
    ASSERT_EQ(Status::OK(), result->status());

    auto z = result->at(0);

    // z->printIndexedBuffer("z");
    // exp.printIndexedBuffer("e");

    ASSERT_TRUE(exp.isSameShape(z));
    ASSERT_TRUE(exp.equalsTo(z));

    delete result;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, sconv2d_1) {
    float _expB[] = {10025.0f,    10350.0f,    10675.0f,    11000.0f,    11325.0f,    11650.0f,    13275.0f,    13600.0f,    13925.0f,    14250.0f,    14575.0f,    14900.0f,    16525.0f,    16850.0f,    17175.0f,    17500.0f,    17825.0f,    18150.0f,    19775.0f,    20100.0f,    20425.0f,    20750.0f,    21075.0f,    21400.0f,    23025.0f,    23350.0f,    23675.0f,    24000.0f,    24325.0f,    24650.0f,    26275.0f,    26600.0f,    26925.0f,    27250.0f,    27575.0f,    27900.0f,    38775.0f,    40350.0f,    41925.0f,    43500.0f,    45075.0f,    46650.0f,    54525.0f,    56100.0f,    57675.0f,    59250.0f,    60825.0f,    62400.0f,    70275.0f,    71850.0f,    73425.0f,    75000.0f,    76575.0f,    78150.0f,    86025.0f,    87600.0f,    89175.0f,    90750.0f,    92325.0f,    93900.0f,   101775.0f,   103350.0f,   104925.0f,    106500.0f,   108075.0f,   109650.0f,   117525.0f,   119100.0f,   120675.0f,   122250.0f,    123825.0f,   125400.0f,    67525.0f,    70350.0f,    73175.0f,    76000.0f,    78825.0f,    81650.0f,    95775.0f,    98600.0f,   101425.0f,   104250.0f,   107075.0f,   109900.0f,    124025.0f,   126850.0f,   129675.0f,   132500.0f,   135325.0f,   138150.0f,   152275.0f,    155100.0f,   157925.0f,   160750.0f,   163575.0f,   166400.0f,   180525.0f,   183350.0f,    186175.0f,   189000.0f,   191825.0f,   194650.0f,   208775.0f,   211600.0f,   214425.0f,    217250.0f,   220075.0f,   222900.0f,   119400.0f,   120350.0f,   121300.0f,   122250.0f,    123200.0f,   124150.0f,   128900.0f,   129850.0f,   130800.0f,   131750.0f,   132700.0f,    133650.0f,   138400.0f,   139350.0f,   140300.0f,   141250.0f,   142200.0f,   143150.0f,    147900.0f,   148850.0f,   149800.0f,   150750.0f,   151700.0f,   152650.0f,   157400.0f,    158350.0f,   159300.0f,   160250.0f,   161200.0f,   162150.0f,   166900.0f,   167850.0f,    168800.0f,   169750.0f,   170700.0f,   171650.0f,   273150.0f,   275350.0f,   277550.0f,    279750.0f,   281950.0f,   284150.0f,   295150.0f,   297350.0f,   299550.0f,   301750.0f,    303950.0f,   306150.0f,   317150.0f,   319350.0f,   321550.0f,   323750.0f,   325950.0f,    328150.0f,   339150.0f,   341350.0f,   343550.0f,   345750.0f,   347950.0f,   350150.0f,    361150.0f,   363350.0f,   365550.0f,   367750.0f,   369950.0f,   372150.0f,   383150.0f,    385350.0f,   387550.0f,   389750.0f,   391950.0f,   394150.0f,   426900.0f,   430350.0f,    433800.0f,   437250.0f,   440700.0f,   444150.0f,   461400.0f,   464850.0f,   468300.0f,    471750.0f,   475200.0f,   478650.0f,   495900.0f,   499350.0f,   502800.0f,   506250.0f,    509700.0f,   513150.0f,   530400.0f,   533850.0f,   537300.0f,   540750.0f,   544200.0f,    547650.0f,   564900.0f,   568350.0f,   571800.0f,   575250.0f,   578700.0f,   582150.0f,    599400.0f,   602850.0f,   606300.0f,   609750.0f,   613200.0f,   616650.0f,    75025.0f,    75350.0f,    75675.0f,    76000.0f,    76325.0f,    76650.0f,    78275.0f,    78600.0f,    78925.0f,    79250.0f,    79575.0f,    79900.0f,    81525.0f,    81850.0f,    82175.0f,    82500.0f,    82825.0f,    83150.0f,    84775.0f,    85100.0f,    85425.0f,    85750.0f,    86075.0f,    86400.0f,    88025.0f,    88350.0f,    88675.0f,    89000.0f,    89325.0f,    89650.0f,    91275.0f,    91600.0f,    91925.0f,    92250.0f,    92575.0f,    92900.0f,    353775.0f,   355350.0f,   356925.0f,   358500.0f,   360075.0f,   361650.0f,   369525.0f,    371100.0f,   372675.0f,   374250.0f,   375825.0f,   377400.0f,   385275.0f,   386850.0f,    388425.0f,   390000.0f,   391575.0f,   393150.0f,   401025.0f,   402600.0f,   404175.0f,    405750.0f,   407325.0f,   408900.0f,   416775.0f,   418350.0f,   419925.0f,   421500.0f,    423075.0f,   424650.0f,   432525.0f,   434100.0f,   435675.0f,   437250.0f,   438825.0f,    440400.0f,   632525.0f,   635350.0f,   638175.0f,   641000.0f,   643825.0f,   646650.0f,    660775.0f,   663600.0f,   666425.0f,   669250.0f,   672075.0f,   674900.0f,   689025.0f,    691850.0f,   694675.0f,   697500.0f,   700325.0f,   703150.0f,   717275.0f,   720100.0f,    722925.0f,   725750.0f,   728575.0f,   73
    Nd4jLong _expS[] = {4, 2, 6, 6, 6, 144, 36, 6, 1, 8192, 1, 99};
    NDArray exp(_expB, _expS);

    int sY = 1;
    int sX = 1;
    int pY = 0;
    int pX = 0;
    int iC = 2;
    int oC = 3;
    int kY = 5;
    int kX = 5;
    int iY = 10;
    int iX = 10;
    int B = 2;

    auto input = NDArrayFactory::create_<float>('c', {B, iC, iY, iX});
    for (int e = 0; e < input->lengthOf(); e++)
        input->p(e, e+1);

    auto weights = NDArrayFactory::create_<float>('c', {oC, iC, kY, kX});
    for (int e = 0; e < weights->lengthOf(); e++)
        weights->p(e, e+1);
    weights->permutei({2,3,1,0});

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, input);
    variableSpace->putVariable(-2, weights);

    auto block = new Context(1, variableSpace, false);
    block->fillInputs({-1, -2});

    block->getIArguments()->push_back(kY);
    block->getIArguments()->push_back(kX);

    block->getIArguments()->push_back(sY);
    block->getIArguments()->push_back(sX);

    block->getIArguments()->push_back(pY);
    block->getIArguments()->push_back(pX);

    // dilation
    block->getIArguments()->push_back(1);
    block->getIArguments()->push_back(1);

    // NOT same mode
    block->getIArguments()->push_back(0);

    nd4j::ops::sconv2d op;

    Nd4jStatus status = op.execute(block);

    ASSERT_EQ(ND4J_STATUS_OK, status);
    auto output = variableSpace->getVariable(1)->getNDArray();

    //exp.printShapeInfo("Expected shape");
    //output->printShapeInfo("Result shape");
    ASSERT_TRUE(exp.isSameShape(output));

        //exp.printBuffer("Expctd buffer");
    //output->printBuffer("Result buffer");
    ASSERT_TRUE(exp.equalsTo(output));

    delete block;
    delete variableSpace;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, sconv2d_2) {
    TypeParam _expBFF[] = {108.9405008,  109.5920008,  110.2435008,  110.8950008,   111.5465008,  112.1980008,  115.4555008,  116.1070008,   116.7585008,  117.410000,   118.061500,   118.7130009,   121.9705009,  122.6220009,  123.2735009,  123.9250009,   124.5765009,  125.2280009,  128.4855009,  129.1370009,   129.7885009,  130.4400009,  131.09150,    131.74300,   135.0005010,  135.6520010,  136.3035010,  136.9550010,   137.6065010,  138.2580010,  141.5155010,  142.1670010,   142.8185010,  143.4700010,  144.1215010,  144.7730010,   248.9617514,  250.670751,   252.3797515,  254.0887515,   255.7977515,  257.5067515,  266.0517515,  267.7607515,   269.469751,   271.1787516,  272.8877516,  274.5967516,   283.1417516,  284.8507516,  286.5597516,  288.268751,   289.9777517,  291.6867517,  300.2317517,  301.9407517,   303.6497517,  305.3587517,  307.067751,   308.7767518,   317.3217518,  319.0307518,  320.7397518,  322.4487518,   324.157751,   325.866751,   334.4117519,  336.1207519,   337.8297519,  339.5387519,  341.2477519,  342.95675,   388.9829964,  391.7494964,  394.5159964,  397.2824964,   400.048996,   402.8154963,  416.647996,   419.4144962,   422.1809962,  424.9474962,  427.7139962,  430.4804962,   444.3129961,  447.0794961,  449.8459961,  452.6124960,   455.3789960,  458.1454960,  471.9779959,  474.7444959,   477.5109959,  480.2774959,  483.0439959,  485.8104958,   499.6429958,  502.4094957,  505.1759957,  507.9424957,   510.7089957,  513.4754957,  527.3079956,  530.0744956,   532.8409956,  535.607495,   538.3739955,  541.1404955,   529.0042487,  532.8282487,  536.6522487,  540.4762487,   544.3002487,  548.1242487,  567.2442487,  571.068248,   574.892248,   578.716248,   582.540248,   586.3642486,   605.4842486,  609.3082486,  613.1322486,  616.9562486,   620.7802486,  624.6042486,  643.7242486,  647.5482486,   651.3722486,  655.1962486,  659.0202486,  662.8442486,   681.9642486,  685.7882486,  689.6122486,  693.4362486,   697.2602486,  701.0842486,  720.2042486,  724.0282486,   727.852248,   731.676248,   735.500248,   739.324248,   669.0255044,  673.9070044,  678.7885044,  683.6700044,   688.5515044,  693.4330044,  717.8405044,  722.7220044,   727.6035044,  732.4850044,  737.3665044,  742.2480044,   766.6555043,  771.5370043,  776.4185043,  781.3000043,   786.1815043,  791.0630043,  815.4705043,  820.3520043,   825.2335043,  830.1150043,  834.9965043,  839.8780043,   864.2855042,  869.1670042,  874.0485042,  878.9300042,   883.8115042,  888.6930042,  913.1005042,  917.9820042,   922.8635042,  927.7450042,  932.6265042,  937.5080042,   809.0467424,  814.9857424,  820.9247424,  826.8637423,   832.8027423,  838.7417423,  868.4367421,  874.3757421,   880.3147420,  886.2537420,  892.1927420,  898.13174,   927.8267418,  933.7657418,  939.7047417,  945.6437417,   951.5827417,  957.5217416,  987.2167415,  993.155741,   999.0947414, 1005.0337414, 1010.972741,  1016.9117413,   1046.6067412, 1052.5457411, 1058.4847411, 1064.4237411,   1070.3627410, 1076.3017410, 1105.996740,  1111.9357408,   1117.8747408, 1123.8137408, 1129.7527407, 1135.6917407,   949.0679815,  956.0644814,  963.060981,   970.0574813,   977.0539812,  984.0504811, 1019.0329807, 1026.0294807,   1033.0259806, 1040.0224805, 1047.0189804, 1054.0154804,   1088.9979800, 1095.9944799, 1102.9909798, 1109.987479,   1116.9839797, 1123.9804796, 1158.9629792, 1165.9594791,   1172.9559791, 1179.9524790, 1186.9489789, 1193.9454788,   1228.9279785, 1235.9244784, 1242.9209783, 1249.9174782,   1256.913978,  1263.9104781, 1298.8929777, 1305.8894776,   1312.8859775, 1319.8824775, 1326.8789774, 1333.8754773,   1089.0892560, 1097.1432561, 1105.1972562, 1113.251256,   1121.3052563, 1129.3592564, 1169.6292568, 1177.6832568,   1185.7372569, 1193.7912570, 1201.845257,  1209.8992571,   1250.1692575, 1258.2232576, 1266.2772576, 1274.3312577,   1282.3852578, 1290.4392579, 1330.7092582, 1338.7632583,   1346.8172584, 1354.8712584, 1362.9252585, 1370.9792586,   1411.24925,   1419.3032590, 1427.3572591, 1435.4112592,   1443.465259,  1451.5192593, 1491.7892597, 1499.8432598,   1507.8972598, 1
    Nd4jLong _expSFF[] = {4, 2, 10, 6, 6, 360, 36, 6, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99,};
    NDArray expFF(_expBFF, _expSFF);

    auto input = NDArrayFactory::create<TypeParam>('c', {2, 3, 10, 10});
    auto weightsD = NDArrayFactory::create<TypeParam>('c', {5, 3, 5, 5});
    auto weightsP = NDArrayFactory::create<TypeParam>('c', {10, 15, 1, 1});

    input.linspace(1);
    weightsD.linspace(1);
    weightsP.linspace(1);
    weightsD.permutei({2,3,1,0});
    weightsP.permutei({2,3,1,0});

    input.applyScalar(scalar::Divide, 100.0);
    weightsD.applyScalar(scalar::Divide, 100.0);
    weightsP.applyScalar(scalar::Divide, 100.0);

    nd4j::ops::sconv2d op;

    auto resultFF = op.execute({&input, &weightsD, &weightsP}, {}, {5, 5, 1, 1, 0, 0, 1, 1, 0, 0}, {});

    auto z = resultFF->at(0);
    //z->printShapeInfo("FF shape");


    ASSERT_TRUE(z->isSameShape(&expFF));

    //expFF.printBuffer("e");
    //z->printBuffer("z");
    ASSERT_TRUE(z->equalsTo(&expFF, 1e-3));

    delete resultFF;
}

TYPED_TEST(TypedConvolutionTests1, sconv2d_3) {
    auto input = NDArrayFactory::create<TypeParam>('c', {3, 3, 8, 8});
    auto weightsD = NDArrayFactory::create<TypeParam>('c', {1, 3, 1, 1});
    auto weightsP = NDArrayFactory::create<TypeParam>('c', {2, 3, 1, 1});
    auto bias = NDArrayFactory::create<TypeParam>('c', {2});
    auto output = NDArrayFactory::create<TypeParam>('c', {3, 2, 8, 8});
    output.assign(0.0);

    input.linspace(1);
    weightsD.linspace(1);
    weightsP.linspace(1);
    bias.linspace(1);
    weightsD.permutei({2,3,1,0});
    weightsP.permutei({2,3,1,0});

    auto expOutput = NDArrayFactory::create<TypeParam>('c', {3, 2, 8, 8});

    nd4j::ops::sconv2d op;
    Nd4jStatus status = op.execute({&input, &weightsD, &weightsP, &bias}, {&output}, {},  {1, 1, 1, 1, 0, 0, 1, 1, 0}, {});
    auto result = op.execute({&input, &weightsD, &weightsP, &bias}, {},  {1, 1, 1, 1, 0, 0, 1, 1, 0}, {});

    auto z = result->at(0);

    //printf("\n");
    //output.printBuffer("output");
    //z->printBuffer("z");


    //ASSERT_TRUE(expOutput.isSameShape(z));

    delete result;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, sconv2d_4) {

    int bS=1, iH=6,iW=6,  iC=3,oC=2,mC=3,  kH=1,kW=1,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=6,oW=6;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iC, iH, iW}, {0.679350, 0.355087, 0.842789, 0.200313, 0.701499, 0.310693, 0.447940, 0.938010, 0.326674, 0.151873, 0.383318, 0.782123, 0.198807,
        0.798564, 0.163263, 0.146968, 0.260897, 0.135058, 0.756209, 0.275454, 0.369088, 0.092826, 0.836492, 0.268413, 0.095062, 0.312795, 0.135918, 0.517544, 0.328703,
        0.061736, 0.396431, 0.248016, 0.548959, 0.115046, 0.814362, 0.721564, 0.404494, 0.299089, 0.403884, 0.988311, 0.022296, 0.927782, 0.318416, 0.068546, 0.284533,
        0.232720, 0.352142, 0.058909, 0.711221, 0.674457, 0.196946, 0.699497, 0.074322, 0.420425, 0.584263, 0.149574, 0.446406, 0.723072, 0.064481, 0.483078, 0.875996,
        0.569819, 0.445863, 0.527755, 0.016646, 0.753678, 0.140636, 0.754129, 0.161932, 0.775037, 0.332645, 0.117394, 0.017711, 0.608476, 0.525152, 0.917194, 0.849891,
        0.589423, 0.852278, 0.390636, 0.889683, 0.669445, 0.698873, 0.961480, 0.157401, 0.157364, 0.493520, 0.569937, 0.126832, 0.115728, 0.786368, 0.737939, 0.490079,
        0.608414, 0.956500, 0.390098, 0.147305, 0.850645, 0.497650, 0.071866, 0.082150, 0.035314, 0.732041, 0.369934, 0.840666, 0.273894, 0.431796, 0.133231});
    NDArray weightsD('c', {kH, kW, iC, mC}, {0.5340641736984253, 0.8257383108139038, 0.3279532492160797, 0.27217748761177063, 0.05432872101664543, 0.31322699785232544, 0.6599581837654114, 0.35526034235954285, 0.5765137672424316});
    NDArray weightsP('c', {1, 1, iC*mC, oC}, {0.4442146420478821, 0.3362849950790405, 0.5215804576873779, 0.5305071473121643, 0.7323054075241089, 0.5168435573577881, 0.8601323962211609, 0.2587810158729553, 0.9473239779472351, 0.39540114998817444, 0.04835261031985283, 0.8724213242530823, 0.8607604503631592, 0.8382210731506348, 0.8573186993598938, 0.6496091485023499, 0.8864102959632874, 0.14267340302467346});
    NDArray biases('c', {1,oC}, {0.8807470202445984, 0.6262521147727966});

    NDArray expOutput('c', {bS, oC, oH, oW}, {1.643804, 2.135067, 2.494167, 2.628944, 2.700440, 2.257452, 2.562539, 2.293667, 2.493985, 2.014933, 2.301736, 2.939066, 1.492952,
        2.026476, 1.771098, 2.013162, 1.315507, 1.289951, 2.831223, 2.196924, 2.028261, 2.024326, 2.983223, 1.809527, 1.434322, 2.513157, 1.826834, 1.608869, 1.297912, 1.212318,
        2.295934, 1.844615, 2.591148, 1.597267, 2.317755, 1.755642, 1.324064, 1.542060, 1.892052, 1.939339, 1.922781, 1.720199, 1.833396, 1.728024, 1.757968, 1.410675, 1.661960,
        2.096277, 1.178815, 1.637460, 1.254187, 1.491076, 0.968625, 0.986342, 2.116042, 1.536920, 1.504321, 1.490398, 2.136795, 1.351860, 1.148578, 1.817408, 1.327139, 1.288620,
        0.962232, 0.980667, 1.623775, 1.417320, 1.845710, 1.237095, 1.762792, 1.352515});

    nd4j::ops::sconv2d op;
    auto results = op.execute({&input, &weightsD, &weightsP, &biases}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

TYPED_TEST(TypedConvolutionTests1, conv2D_BP_Bias_1) {
    TypeParam _expWGradB[] = {9312.0, 12580.0, 9528.0, 13168.0, 17712.0, 13360.0, 9960.0, 13348.0, 10032.0, 13344.0, 18148.0, 13848.0, 19312.0, 26160.0, 19888.0, 15144.0, 20452.0, 15504.0};
    Nd4jLong _expWGradS[] = {4, 2, 1, 3, 3, 9, 9, 3, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99};
    NDArray expWGrad(_expWGradB, _expWGradS);
    expWGrad.permutei({2,3,1,0});

    TypeParam _expBGradB[] = {784.0, 1296.0};
    Nd4jLong _expBGradS[] = {2, 2, 1, 1, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99};

    NDArray expBGrad(_expBGradB, _expBGradS);

    auto input = NDArrayFactory::create<TypeParam>('c', {2, 1, 4, 4});
    auto weights = NDArrayFactory::create<TypeParam>('c', {2, 1, 3, 3});
    auto bias = NDArrayFactory::create<TypeParam>('c', {2, 1});
    auto epsilonNext = NDArrayFactory::create<TypeParam>('c', {2, 2, 4, 4});


    TypeParam _expEpsB[] = {952.0, 1540.0, 1636.0, 1180.0, 1791.0, 2886.0, 3057.0, 2193.0, 2223.0, 3570.0, 3741.0, 2673.0, 1900.0, 3028.0, 3160.0, 2240.0, 2872.0, 4612.0, 4708.0, 3356.0, 5247.0, 8358.0, 8529.0, 6033.0, 5679.0, 9042.0, 9213.0, 6513.0, 4588.0, 7252.0, 7384.0, 5184.0};
    NDArray expEps(_expEpsB, input.getShapeInfo());

    input.linspace(1);
    weights.linspace(1);
    epsilonNext.linspace(1);
    weights.permutei({2,3,1,0});

    nd4j::ops::conv2d_bp op;

    auto results = op.execute({&input, &weights, &bias, &epsilonNext}, {},  {3, 3, 1, 1, 0, 0, 1, 1, 1}, {});

    ASSERT_TRUE(results->size() == 3);

    auto epsilon = results->at(0);
    auto gradW = results->at(1);
    auto gradB = results->at(2);

    ASSERT_TRUE(expWGrad.isSameShape(gradW));

    //expWGrad.printBuffer("Expctd buffer");
    //  gradW->printBuffer("Result buffer");
    ASSERT_TRUE(expWGrad.equalsTo(gradW));


    ASSERT_TRUE(input.isSameShape(epsilon));

    //  expEps.printBuffer("Expctd buffer");
    //epsilon->printBuffer("Result buffer");
    ASSERT_TRUE(expEps.equalsTo(epsilon));

    ASSERT_TRUE(expBGrad.isSameShape(gradB));

    ASSERT_TRUE(expBGrad.equalsTo(gradB));

    delete results;
}


TYPED_TEST(TypedConvolutionTests1, conv2D_BP_NoBias_1) {
    TypeParam _expWGradB[] = {9312.0, 12580.0, 9528.0, 13168.0, 17712.0, 13360.0, 9960.0, 13348.0, 10032.0, 13344.0, 18148.0, 13848.0, 19312.0, 26160.0, 19888.0, 15144.0, 20452.0, 15504.0};
    Nd4jLong _expWGradS[] = {4, 2, 1, 3, 3, 9, 9, 3, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99};
    NDArray expWGrad(_expWGradB, _expWGradS);
    expWGrad.permutei({2,3,1,0});

    auto input = NDArrayFactory::create<TypeParam>('c', {2, 1, 4, 4});
    auto weights = NDArrayFactory::create<TypeParam>('c', {2, 1, 3, 3});
    auto epsilonNext = NDArrayFactory::create<TypeParam>('c', {2, 2, 4, 4});


    TypeParam _expEpsB[] = {952.0, 1540.0, 1636.0, 1180.0, 1791.0, 2886.0, 3057.0, 2193.0, 2223.0, 3570.0, 3741.0, 2673.0, 1900.0, 3028.0, 3160.0, 2240.0, 2872.0, 4612.0, 4708.0, 3356.0, 5247.0, 8358.0, 8529.0, 6033.0, 5679.0, 9042.0, 9213.0, 6513.0, 4588.0, 7252.0, 7384.0, 5184.0};
    NDArray expEps(_expEpsB, input.getShapeInfo());

    input.linspace(1);
    weights.linspace(1);
    epsilonNext.linspace(1);
    weights.permutei({2,3,1,0});

    nd4j::ops::conv2d_bp op;

    auto results = op.execute({&input, &weights, &epsilonNext}, {},  {3, 3, 1, 1, 0, 0, 1, 1, 1}, {});

    ASSERT_TRUE(results->size() == 2);

    auto epsilon = results->at(0);
    auto gradW = results->at(1);

    ASSERT_TRUE(expWGrad.isSameShape(gradW));

    //expWGrad.printBuffer("Expctd buffer");
    //  gradW->printBuffer("Result buffer");
    ASSERT_TRUE(expWGrad.equalsTo(gradW));


    ASSERT_TRUE(input.isSameShape(epsilon));

    //  expEps.printBuffer("Expctd buffer");
    //epsilon->printBuffer("Result buffer");
    ASSERT_TRUE(expEps.equalsTo(epsilon));

    delete results;
}

TYPED_TEST(TypedConvolutionTests1, sconv2d_conv2d_1) {
    TypeParam _expBFF[] = {10025.0f,   10350.0f,   10675.0f,   11000.0f,   11325.0f,   11650.0f,   13275.0f,   13600.0f,   13925.0f,   14250.0f,   14575.0f,   14900.0f,   16525.0f,   16850.0f,   17175.0f,   17500.0f,   17825.0f,   18150.0f,   19775.0f,   20100.0f,   20425.0f,   20750.0f,   21075.0f,   21400.0f,   23025.0f,   23350.0f,   23675.0f,   24000.0f,   24325.0f,   24650.0f,   26275.0f,   26600.0f,   26925.0f,   27250.0f,   27575.0f,   27900.0f,   53150.0f,   55350.0f,   57550.0f,   59750.0f,   61950.0f,   64150.0f,   75150.0f,   77350.0f,   79550.0f,   81750.0f,   83950.0f,   86150.0f,   97150.0f,   99350.0f,  101550.0f,  103750.0f,  105950.0f,  108150.0f,   119150.0f,  121350.0f,  123550.0f,  125750.0f,  127950.0f,  130150.0f,   141150.0f,  143350.0f,  145550.0f,  147750.0f,  149950.0f,  152150.0f,   163150.0f,  165350.0f,  167550.0f,  169750.0f,  171950.0f,  174150.0f,   119400.0f,  120350.0f,  121300.0f,  122250.0f,  123200.0f,  124150.0f,   128900.0f,  129850.0f,  130800.0f,  131750.0f,  132700.0f,  133650.0f,   138400.0f,  139350.0f,  140300.0f,  141250.0f,  142200.0f,  143150.0f,   147900.0f,  148850.0f,  149800.0f,  150750.0f,  151700.0f,  152650.0f,   157400.0f,  158350.0f,  159300.0f,  160250.0f,  161200.0f,  162150.0f,   166900.0f,  167850.0f,  168800.0f,  169750.0f,  170700.0f,  171650.0f,   350025.0f,  352850.0f,  355675.0f,  358500.0f,  361325.0f,  364150.0f,   378275.0f,  381100.0f,  383925.0f,  386750.0f,  389575.0f,  392400.0f,   406525.0f,  409350.0f,  412175.0f,  415000.0f,  417825.0f,  420650.0f,   434775.0f,  437600.0f,  440425.0f,  443250.0f,  446075.0f,  448900.0f,   463025.0f,  465850.0f,  468675.0f,  471500.0f,  474325.0f,  477150.0f,   491275.0f,  494100.0f,  496925.0f,  499750.0f,  502575.0f,  505400.0f,   353775.0f,  355350.0f,  356925.0f,  358500.0f,  360075.0f,  361650.0f,   369525.0f,  371100.0f,  372675.0f,  374250.0f,  375825.0f,  377400.0f,   385275.0f,  386850.0f,  388425.0f,  390000.0f,  391575.0f,  393150.0f,   401025.0f,  402600.0f,  404175.0f,  405750.0f,  407325.0f,  408900.0f,   416775.0f,  418350.0f,  419925.0f,  421500.0f,  423075.0f,  424650.0f,   432525.0f,  434100.0f,  435675.0f,  437250.0f,  438825.0f,  440400.0f,   771900.0f,  775350.0f,  778800.0f,  782250.0f,  785700.0f,  789150.0f,   806400.0f,  809850.0f,  813300.0f,  816750.0f,  820200.0f,  823650.0f,   840900.0f,  844350.0f,  847800.0f,  851250.0f,  854700.0f,  858150.0f,   875400.0f,  878850.0f,  882300.0f,  885750.0f,  889200.0f,  892650.0f,   909900.0f,  913350.0f,  916800.0f,  920250.0f,  923700.0f,  927150.0f,   944400.0f,  947850.0f,  951300.0f,  954750.0f,  958200.0f,  961650.0f,   107525.0f,  107850.0f,  108175.0f,  108500.0f,  108825.0f,  109150.0f,   110775.0f,  111100.0f,  111425.0f,  111750.0f,  112075.0f,  112400.0f,   114025.0f,  114350.0f,  114675.0f,  115000.0f,  115325.0f,  115650.0f,   117275.0f,  117600.0f,  117925.0f,  118250.0f,  118575.0f,  118900.0f,   120525.0f,  120850.0f,  121175.0f,  121500.0f,  121825.0f,  122150.0f,   123775.0f,  124100.0f,  124425.0f,  124750.0f,  125075.0f,  125400.0f,   713150.0f,  715350.0f,  717550.0f,  719750.0f,  721950.0f,  724150.0f,   735150.0f,  737350.0f,  739550.0f,  741750.0f,  743950.0f,  746150.0f,   757150.0f,  759350.0f,  761550.0f,  763750.0f,  765950.0f,  768150.0f,   779150.0f,  781350.0f,  783550.0f,  785750.0f,  787950.0f,  790150.0f,   801150.0f,  803350.0f,  805550.0f,  807750.0f,  809950.0f,  812150.0f,   823150.0f,  825350.0f,  827550.0f,  829750.0f,  831950.0f,  834150.0f,   404400.0f,  405350.0f,  406300.0f,  407250.0f,  408200.0f,  409150.0f,   413900.0f,  414850.0f,  415800.0f,  416750.0f,  417700.0f,  418650.0f,   423400.0f,  424350.0f,  425300.0f,  426250.0f,  427200.0f,  428150.0f,   432900.0f,  433850.0f,  434800.0f,  435750.0f,  436700.0f,  437650.0f,   442400.0f,  443350.0f,  444300.0f,  445250.0f,  446200.0f,  447150.0f,   451900.0f,  452850.0f,  453800.0f,  454750.0f,  455700.0f,  456650.0f,   1197525.0f, 1200350.0f, 1203175.0f, 1206000.0f, 1208825.0f, 1211650.0f,   1225775.0f, 1228600.0f, 1231425.0f, 1234250.0f, 1237075.0f, 1
    Nd4jLong _expSFF[] = {4, 2, 6, 6, 6, 216, 36, 6, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99,};
    NDArray expFF(_expBFF, _expSFF);
    TypeParam _exp2BFF[] = {827.4900282f,   832.2350283f,   836.9800284f,   841.725028f,   846.4700287f,   851.2150288f,   874.9400293f,   879.6850294f,   884.4300295f,   889.1750296f,   893.9200297f,   898.665029f,   922.3900304f,   927.1350305f,   931.8800306f,   936.6250307f,   941.3700308f,   946.1150309f,   969.8400315f,   974.5850316f,   979.3300317f,   984.0750318f,   988.8200319f,   993.5650320f,   1017.2900326f,  1022.0350327f,  1026.7800328f,  1031.5250329f,   1036.2700330f,  1041.0150331f,  1064.7400337f,  1069.4850338f,   1074.2300339f,  1078.9750340f,  1083.7200341f,  1088.4650342f,   1822.4550553f,  1833.995055f,   1845.5350558f,  1857.075056f,   1868.6150563f,  1880.1550566f,  1937.8550578f,  1949.3950581f,   1960.9350583f,  1972.4750586f,  1984.015058f,   1995.5550591f,   2053.2550604f,  2064.7950606f,  2076.3350609f,  2087.8750611f,   2099.4150614f,  2110.955061f,   2168.6550629f,  2180.1950632f,   2191.7350634f,  2203.2750637f,  2214.8150639f,  2226.3550642f,   2284.0550655f,  2295.5950657f,  2307.1350660f,  2318.6750662f,   2330.2150665f,  2341.7550667f,  2399.4550680f,  2410.9950683f,   2422.5350685f,  2434.0750688f,  2445.6150690f,  2457.1550693f,   2817.419968f,   2835.7549686f,  2854.0899683f,  2872.4249680f,   2890.7599677f,  2909.0949674f,  3000.7699660f,  3019.104965f,   3037.4399655f,  3055.7749652f,  3074.1099649f,  3092.4449646f,   3184.1199632f,  3202.4549629f,  3220.789962f,   3239.1249624f,   3257.4599621f,  3275.7949618f,  3367.4699604f,  3385.8049601f,   3404.1399598f,  3422.474959f,   3440.8099593f,  3459.1449590f,   3550.8199576f,  3569.1549573f,  3587.4899570f,  3605.8249567f,   3624.1599565f,  3642.4949562f,  3734.1699548f,  3752.5049545f,   3770.8399542f,  3789.1749539f,  3807.5099536f,  3825.8449534f,   3812.385098f,   3837.5150988f,  3862.6450994f,  3887.7751000f,   3912.9051006f,  3938.0351012f,  4063.6851041f,  4088.8151047f,   4113.9451053f,  4139.0751059f,  4164.2051065f,  4189.3351071f,   4314.9851100f,  4340.1151106f,  4365.2451112f,  4390.3751118f,   4415.5051124f,  4440.6351130f,  4566.2851159f,  4591.4151165f,   4616.5451171f,  4641.6751177f,  4666.805118f,   4691.9351188f,   4817.5851218f,  4842.7151224f,  4867.8451230f,  4892.975123f,   4918.1051241f,  4943.2351247f,  5068.8851277f,  5094.0151283f,   5119.1451288f,  5144.2751294f,  5169.4051300f,  5194.5351306f,   4807.3499803f,  4839.2749801f,  4871.1999799f,  4903.1249797f,   4935.0499795f,  4966.9749793f,  5126.5999784f,  5158.5249782f,   5190.4499780f,  5222.3749778f,  5254.2999777f,  5286.2249775f,   5445.8499765f,  5477.774976f,   5509.6999762f,  5541.6249760f,   5573.5499758f,  5605.4749756f,  5765.0999747f,  5797.0249745f,   5828.9499743f,  5860.8749741f,  5892.7999739f,  5924.724973f,   6084.3499728f,  6116.2749726f,  6148.1999724f,  6180.1249723f,   6212.0499721f,  6243.9749719f,  6403.59997f,    6435.5249708f,   6467.4499706f,  6499.3749704f,  6531.2999702f,  6563.2249700f,   5802.3150007f,  5841.0350006f,  5879.7550005f,  5918.4750004f,   5957.195000f,   5995.9150003f,  6189.5149999f,  6228.2349998f,   6266.9549997f,  6305.6749996f,  6344.3949995f,  6383.114999f,   6576.7149990f,  6615.4349990f,  6654.1549989f,  6692.8749988f,   6731.5949987f,  6770.3149986f,  6963.9149982f,  7002.6349981f,   7041.3549981f,  7080.0749980f,  7118.7949979f,  7157.5149978f,   7351.1149974f,  7389.8349973f,  7428.5549972f,  7467.2749972f,   7505.9949971f,  7544.7149970f,  7738.3149966f,  7777.0349965f,   7815.7549964f,  7854.4749963f,  7893.1949963f,  7931.9149962f,   6797.2799488f,  6842.794948f,   6888.3099489f,  6933.8249490f,   6979.3399491f,  7024.8549492f,  7252.4299497f,  7297.9449498f,   7343.4599499f,  7388.9749500f,  7434.489950f,   7480.0049501f,   7707.5799506f,  7753.0949507f,  7798.6099508f,  7844.1249509f,   7889.6399510f,  7935.1549511f,  8162.7299515f,  8208.2449516f,   8253.7599517f,  8299.2749518f,  8344.7899519f,  8390.3049520f,   8617.8799525f,  8663.394952f,   8708.9099526f,  8754.4249527f,   8799.9399528f,  8845.4549529f,  9073.0299534f,  9118.5449535f,   9164.0599536f,  9209.5749537f,  9255.089953f,   93
    Nd4jLong _exp2SFF[] = {4, 2, 10, 6, 6, 360, 36, 6, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99,};
    NDArray exp2FF(_exp2BFF, _exp2SFF);

    auto input = NDArrayFactory::create<TypeParam>('c', {2, 3, 10, 10});
    auto weightsD = NDArrayFactory::create<TypeParam>('c', {2, 3, 5, 5});
    auto weightsP = NDArrayFactory::create<TypeParam>('c', {10, 6, 1, 1});


    input.linspace(1);
    weightsD.linspace(1);
    weightsP.linspace(1);
    weightsD.permutei({2,3,1,0});
    weightsP.permutei({2,3,1,0});

    weightsP.applyScalar(scalar::Divide, 10000.0);

    nd4j::ops::sconv2d op;
    auto resultFF = op.execute({&input, &weightsD}, {}, {5, 5, 1, 1, 0, 0, 1, 1, 0}, {});

    auto z = resultFF->at(0);

    ASSERT_TRUE(z->isSameShape(&expFF));
    ASSERT_TRUE(z->equalsTo(&expFF, 1));


    nd4j::ops::conv2d op2d;
    // weightsP.printShapeInfo();
    auto result2D = op2d.execute({z, &weightsP}, {}, {1, 1, 1, 1, 0, 0, 1, 1, 0, 0}, {});

    auto z2d = result2D->at(0);

    ASSERT_TRUE(z2d->isSameShape(&exp2FF));
    ASSERT_TRUE(z2d->equalsTo(&exp2FF));

    delete resultFF;
    delete result2D;
}

TEST_F(ConvolutionTests1, TestDeconv_bp_1) {

    int bS=3, iH=4,iW=4,  iC=3,oC=2,  kH=1,kW=1,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=4,oW=4;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW


    NDArray input('c', {bS, iC, iH, iW}, nd4j::DataType::FLOAT32);
    NDArray bias('c', {oC}, nd4j::DataType::FLOAT32);
    NDArray weights('c',{kH,kW,oC,iC}, {1,3,5,2,4,6}, nd4j::DataType::FLOAT32);
    NDArray gradO('c', {bS, oC, oH, oW},nd4j::DataType::FLOAT32);

    NDArray expGradI('c', {bS, iC, iH, iW}, {35.f,   38.f,   41.f,   44.f,   47.f,   50.f,   53.f,   56.f,   59.f,   62.f,   65.f,    68.f,   71.f,   74.f,
                77.f,   80.f,   71.f,   78.f,   85.f,   92.f,   99.f,  106.f,    113.f,  120.f,  127.f,  134.f,  141.f,  148.f,  155.f,  162.f,  169.f,
                176.f,  107.f,    118.f,  129.f,  140.f,  151.f,  162.f,  173.f,  184.f,  195.f,  206.f,  217.f,  228.f,    239.f,  250.f,  261.f,  272.f,
                131.f,  134.f,  137.f,  140.f,  143.f,  146.f,  149.f,    152.f,  155.f,  158.f,  161.f,  164.f,  167.f,  170.f,  173.f,  176.f,  295.f,
                302.f,    309.f,  316.f,  323.f,  330.f,  337.f,  344.f,  351.f,  358.f,  365.f,  372.f,  379.f,    386.f,  393.f,  400.f,  459.f,  470.f,
                481.f,  492.f,  503.f,  514.f,  525.f,  536.f,    547.f,  558.f,  569.f,  580.f,  591.f,  602.f,  613.f,  624.f,  227.f,  230.f,  233.f,
                236.f,  239.f,  242.f,  245.f,  248.f,  251.f,  254.f,  257.f,  260.f,  263.f,  266.f,    269.f,  272.f,  519.f,  526.f,  533.f,  540.f,
                547.f,  554.f,  561.f,  568.f,  575.f,    582.f,  589.f,  596.f,  603.f,  610.f,  617.f,  624.f,  811.f,  822.f,  833.f,  844.f,    855.f,
                866.f,  877.f,  888.f,  899.f,  910.f,  921.f,  932.f,  943.f,  954.f,  965.f,    976.f}, nd4j::DataType::FLOAT32);
    NDArray expGradW('c', {kH, kW, oC, iC}, {160008., 191112., 222216., 203400., 246792., 290184.f}, nd4j::DataType::FLOAT32);
    NDArray expGradB('c', {oC}, {1944.f,  2712.f}, nd4j::DataType::FLOAT32);

    input.linspace(1);
    bias.linspace(1);
    gradO.linspace(1);


    nd4j::ops::deconv2d_bp op;
    auto results = op.execute({&input, &weights, &bias, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});

    ASSERT_EQ(ND4J_STATUS_OK, results->status());

    auto gradI = results->at(0);
    auto gradW = results->at(1);
    auto gradB = results->at(2);

    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    ASSERT_TRUE(expGradW.isSameShape(gradW));
    ASSERT_TRUE(expGradW.equalsTo(gradW));

    ASSERT_TRUE(expGradB.isSameShape(gradB));
    ASSERT_TRUE(expGradB.equalsTo(gradB));

    delete results;
}
TEST_F(ConvolutionTests1, TestDeconv_bp_2) {
    /*
     Input shape:
    [3, 3, 14, 14]
    Output shape:
    [3, 2, 15, 15]
    Weights shape:
    [3, 2, 2, 2]
    Bias shape:
    [1, 2]
    weight shape:
    [3, 2, 2, 2]
    weight grad shape:
    [3, 2, 2, 2]
    bias grad shape:
    [2]
    input epsilon shape:
    [3, 2, 15, 15]
    output epsilon shape:
    [3, 3, 14, 14]
     */
    /*
    auto input('c', {3, 3, 14, 14});
    auto bias('c', {2});
    auto weights('c',{3, 2, 2, 2});
    auto epsilon('c', {3, 2, 15, 15});


    input.linspace(1);
    weights.linspace(1);
    bias.linspace(1);
    epsilon.linspace(1);

    nd4j::ops::deconv2d_bp<double> op;

    auto result = op.execute({&input, &weights, &bias, &epsilon}, {}, {2, 2, 1, 1, 0, 0, 2, 2, 0});
    ASSERT_EQ(ND4J_STATUS_OK, result->status());


    delete result;*/
}
TYPED_TEST(TypedConvolutionTests1, Test_Conv1D_ff_1) {
    auto input = NDArrayFactory::create<TypeParam>('c', {2, 2, 6});
    auto weights = NDArrayFactory::create<TypeParam>('c', {2, 2, 3}, {1,5,9,3,7,11,2,6,10,4,8,12});
    auto bias = NDArrayFactory::create<TypeParam>('c', {3});
    auto expFF = NDArrayFactory::create<TypeParam>('c', {2, 3, 5}, {59.0f, 69.0f, 79.0f, 89.0f, 99.0f, 132.0f, 158.0f, 184.0f, 210.0f, 236.0f, 205.0f, 247.0f, 289.0f, 331.0f, 373.0f, 179.0f, 189.0f, 199.0f, 209.0f, 219.0f, 444.0f, 470.0f, 496.0f, 522.0f, 548.0f, 709.0f, 751.0f, 793.0f, 835.0f, 877.0f});
    auto expEps = NDArrayFactory::create<TypeParam>('c', {2, 2, 6}, {130.0f, 293.0f, 326.0f, 359.0f, 392.0f, 220.0f, 166.0f, 371.0f, 416.0f, 461.0f, 506.0f, 280.0f, 355.0f, 788.0f, 821.0f, 854.0f, 887.0f, 490.0f, 481.0f, 1046.0f, 1091.0f, 1136.0f, 1181.0f, 640.0f});
    auto expGW = NDArrayFactory::create<TypeParam>('c', {3, 2, 2}, {1415.0f, 1520.0f, 2045.0f, 2150.0f, 1865.0f, 2020.0f, 2795.0f, 2950.0f, 2315.0f, 2520.0f, 3545.0f, 3750.0f});
    auto expGB = NDArrayFactory::create<TypeParam>('c', {3}, {105.0f, 155.0f, 205.0f});

    expGW.permutei({2,1,0});
    input.linspace(1);
    bias.linspace(1);

    nd4j::ops::conv1d op;
    auto result_FF = op.execute({&input, &weights, &bias}, {}, {2, 1, 0, 1, 0, 0});

    ASSERT_EQ(ND4J_STATUS_OK, result_FF->status());

    auto z = result_FF->at(0);

    ASSERT_TRUE(expFF.isSameShape(z));
    ASSERT_TRUE(expFF.equalsTo(z));

    nd4j::ops::conv1d_bp op_bp;

    auto epsilonNxt = z->dup();
    epsilonNxt->linspace(1);

    auto result_BP = op_bp.execute({&input, &weights, &bias, epsilonNxt}, {}, {2, 1, 0, 1, 0, 0});
    ASSERT_EQ(ND4J_STATUS_OK, result_BP->status());

    auto eps = result_BP->at(0);
    auto gradW = result_BP->at(1);
    auto gradB = result_BP->at(2);

    ASSERT_TRUE(expEps.isSameShape(eps));
    ASSERT_TRUE(expGW.isSameShape(gradW));
    ASSERT_TRUE(expGB.isSameShape(gradB));

    ASSERT_TRUE(expEps.equalsTo(eps));
    ASSERT_TRUE(expGW.equalsTo(gradW));
    ASSERT_TRUE(expGB.equalsTo(gradB));

    delete result_FF;
    delete result_BP;
    delete epsilonNxt;
}


TYPED_TEST(TypedConvolutionTests1, Test_Conv1D_ff_2) {
    auto input = NDArrayFactory::create<TypeParam>('c', {2, 2, 6});
    auto weights = NDArrayFactory::create<TypeParam>('c', {2, 2, 3}, {1.f, 5.f, 9.f, 3.f, 7.f, 11.f, 2.f, 6.f, 10.f, 4.f, 8.f, 12.f});

    input.linspace(1);

    nd4j::ops::conv1d op;
    auto result = op.execute({&input, &weights}, {}, {2, 1, 0, 1, 1,0});

    ASSERT_EQ(ND4J_STATUS_OK, result->status());

    auto z = result->at(0);

    delete result;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, conv1d_causal_1) {

    int bS=2, iW=3,  iC=4,oC=3,  kW=2,  sW=1,  pW=0,  dW=1;
    int oW = (iW-1)/sW + 1;
    int paddingMode = 2;             // CAUSAL
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iW, iC});
    NDArray weights('c', {kW, iC, oC});
    NDArray bias('c', {oC}, {-1,-2,-3});

    NDArray expOutput('c', {bS, oW, oC}, {18. ,  18. ,  18. , 53. ,  55.6,  58.2, 89.8,  95.6, 101.4, 102. , 106.8, 111.6, 163.4, 175.6, 187.8, 200.2, 215.6, 231.});

    input.linspace(1., 1.);
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv1d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kW, sW, pW, dW,  paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, conv1d_causal_2) {

    int bS=2, iW=16,  iC=3,oC=4,  kW=2,  sW=2,  pW=0,  dW=1;
    int oW = (iW-1)/sW + 1;
    int paddingMode = 2;             // CAUSAL
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iW, iC});
    NDArray weights('c', {kW, iC, oC});
    NDArray bias('c', {oC}, {-1,-2,-3,-4});

    NDArray expOutput('c', {bS, oW, oC}, { 10. ,   9.6,   9.2,   8.8, 48.9,  51.8,  54.7,  57.6, 88.5,  95. , 101.5, 108. , 128.1, 138.2, 148.3, 158.4,
                                           167.7, 181.4, 195.1, 208.8, 207.3, 224.6, 241.9, 259.2, 246.9, 267.8, 288.7, 309.6, 286.5, 311. , 335.5, 360. ,
                                           254.8, 268.8, 282.8, 296.8, 365.7, 397.4, 429.1, 460.8, 405.3, 440.6, 475.9, 511.2, 444.9, 483.8, 522.7, 561.6,
                                           484.5, 527. , 569.5, 612. , 524.1, 570.2, 616.3, 662.4, 563.7, 613.4, 663.1, 712.8, 603.3, 656.6, 709.9, 763.2});

    input.linspace(1., 1.);
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv1d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kW, sW, pW, dW, paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, conv1d_causal_3) {

    int bS=2, iW=16,  iC=3,oC=4,  kW=3,  sW=3,  pW=0,  dW=1;
    int oW = (iW-1)/sW + 1;
    int paddingMode = 2;             // CAUSAL
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iW, iC});
    NDArray weights('c', {kW, iC, oC});
    NDArray bias('c', {oC}, {-1,-2,-3,-4});

    NDArray expOutput('c', {bS, oW, oC}, {17.2,   16.8,   16.4,   16.,145.4,  151.6,  157.8,  164.,283.1,  297.4,  311.7,  326.,  420.8,  443.2,  465.6,  488.,
                                558.5,  589.,  619.5,  650.,696.2001,  734.8,  773.4,  812.,  434.8,  448.8,  462.8,  476.8,   879.8,  929.2,  978.6, 1028.,
                                1017.5, 1075., 1132.5, 1190.,1155.2001, 1220.8, 1286.4, 1352.,1292.8999, 1366.6, 1440.3, 1514.,  1430.6001, 1512.4, 1594.2, 1676.});

    input.linspace(1., 1.);
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv1d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kW, sW, pW, dW, paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, conv1d_causal_4) {

    int bS=2, iW=8,  iC=3,oC=4,  kW=3,  sW=1,  pW=0,  dW=3;
    int oW = (iW-1)/sW + 1;
    int paddingMode = 2;             // CAUSAL
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iW, iC});
    NDArray weights('c', {kW, iC, oC});
    NDArray bias('c', {oC}, {-1,-2,-3,-4});

    NDArray expOutput('c', {bS, oW, oC}, {17.2,  16.8,  16.4,  16. ,43.3,  43.8,  44.3,  44.8,69.4,  70.8,  72.2,  73.6,106.5, 109.4, 112.3, 115.2,147.9, 152.6, 157.3, 162. ,189.3, 195.8, 202.3,
                                        208.8,234.5, 243.4, 252.3, 261.2,280.4, 292. , 303.6, 315.2, 226. , 232.8, 239.6, 246.4,  252.1, 259.8, 267.5, 275.2,278.2, 286.8, 295.4, 304. ,437.7,
                                        455. , 472.3, 489.6,479.1, 498.2, 517.3, 536.4,520.5, 541.4, 562.3, 583.2,  601.7, 632.2, 662.7, 693.2, 647.6, 680.8, 714. , 747.2});

    input.linspace(1., 1.);
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv1d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kW, sW, pW, dW, paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, conv1d_causal_5) {

    int bS=2, iW=8,  iC=3,oC=4,  kW=3,  sW=1,  pW=0,  dW=3;
    int oW = (iW-1)/sW + 1;
    int paddingMode = 2;             // CAUSAL
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iC, iW});
    NDArray weights('c', {kW, iC, oC});
    NDArray bias('c', {oC}, {-1,-2,-3,-4});

    NDArray expOutput('c', {bS, oC, oW}, { 83.7,  92.4, 101.1, 162.1, 175.9, 189.7, 223.4, 238.7,85.4,  94.4, 103.4, 167.4, 181.8, 196.2, 233.2, 249.4,87.1,  96.4, 105.7, 172.7, 187.7, 202.7, 243. , 260.1,
                         88.8,  98.4, 108. , 178. , 193.6, 209.2, 252.8, 270.8, 292.5, 301.2, 309.9, 493.3, 507.1, 520.9, 590.6, 605.9,  301.4, 310.4, 319.4, 513. , 527.4, 541.8, 622. , 638.2,
                        310.3, 319.6, 328.9, 532.7, 547.7, 562.7, 653.4, 670.5,  319.2, 328.8, 338.4, 552.4, 568. , 583.6, 684.8, 702.8});

    input.linspace(1., 1.);
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv1d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kW, sW, pW, dW, paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, conv1d_causal_6) {

    int bS=2, iW=16,  iC=3,oC=4,  kW=3,  sW=3,  pW=0,  dW=1;
    int oW = (iW-1)/sW + 1;
    int paddingMode = 2;             // CAUSAL
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iC, iW});
    NDArray weights('c', {kW, iC, oC});
    NDArray bias('c', {oC}, {-1,-2,-3,-4});

    NDArray expOutput('c', {bS, oC, oW}, {159.7,335.3,381.2,427.1,473. ,518.9,163.8,351.4,400. ,448.6,497.2,545.8,167.9,367.5,418.8,470.1,521.4,572.7,172. ,383.6,437.6,491.6,545.6,599.6,
                                577.3, 1069.7, 1115.6, 1161.5, 1207.4, 1253.3,595.8, 1129. , 1177.6, 1226.2, 1274.8, 1323.4,614.3, 1188.3, 1239.6, 1290.9, 1342.2, 1393.5,
                                632.8, 1247.6, 1301.6, 1355.6, 1409.6, 1463.6});

    input.linspace(1., 1.);
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv1d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kW, sW, pW, dW, paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, conv1d_causal_bp_1) {

    int bS=2, iW=3,  iC=4,oC=3,  kW=2,  sW=1,  pW=0,  dW=1;
    int oW = (iW-1)/sW + 1;
    int paddingMode = 2;             // CAUSAL
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iW, iC});
    NDArray weights('c', {kW, iC, oC});
    NDArray bias('c', {oC}, {-1,-2,-3});
    NDArray gradO('c', {bS, oW, oC});

    input.linspace(1., 1.);
    weights.linspace(0.1, 0.1);
    gradO.linspace(-1.5, 0.1);

    const OpArgsHolder argsHolderFF({&input, &weights, &bias}, {}, {kW, sW, pW, dW,  paddingMode, dataFormat});
    const OpArgsHolder argsHolderBP({&input, &weights, &bias, &gradO}, {}, {kW, sW, pW, dW,  paddingMode, dataFormat});

    nd4j::ops::conv1d opFF;
    nd4j::ops::conv1d_bp opBP;

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

    ASSERT_TRUE(isGradCorrect);
}

TEST_F(ConvolutionTests1, Test_Dilation2D_1) {
    auto input = NDArrayFactory::create<double>('c', {2, 6, 6, 3});
    auto weights = NDArrayFactory::create<double>('c', {3, 2, 3});
    auto exp = NDArrayFactory::create<double>('c', {2, 3, 3, 3}, {77,   79,   81,   83,   85,   87,   80,   82,   84,  113,  115,  117, 119,  121,  123,  116,  118,  120,  107,  109,  111,  113,  115,  117, 110,  112,  114,  185,  187,  189,  191,  193,  195,  188,  190,  192, 221,  223,  225,  227,  229,  231,  224,  226,  228,  215,  217,  219, 221,  223,  225,  218,  220,  222,});

    input.linspace(1);
    weights.linspace(1);

    nd4j::ops::dilation2d op;
    auto result = op.execute({&input, &weights}, {}, {1, 1,2,2,1, 1,2,2,1});
    ASSERT_EQ(Status::OK(), result->status());

    auto z = result->at(0);

    ASSERT_TRUE(exp.isSameShape(z));
    ASSERT_TRUE(exp.equalsTo(z));

    delete result;
}

TEST_F(ConvolutionTests1, Test_Dilation2D_2) {
    auto input = NDArrayFactory::create<double>('c', {2, 6, 6, 3});
    auto weights = NDArrayFactory::create<double>('c', {3, 2, 3});
    auto exp = NDArrayFactory::create<double>('c', {2, 1, 2, 3}, {95, 97, 99, 101, 103, 105, 203, 205, 207, 209, 211, 213});

    input.linspace(1);
    weights.linspace(1);

    nd4j::ops::dilation2d op;
    auto result = op.execute({&input, &weights}, {}, {0, 1,2,2,1, 1,2,2,1});
    ASSERT_EQ(Status::OK(), result->status());

    auto z = result->at(0);

    ASSERT_TRUE(exp.isSameShape(z));
    ASSERT_TRUE(exp.equalsTo(z));

    delete result;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_bp_test1) {

    int bS=2, iH=4,iW=3,  iC=4,oC=3,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=4,oW=3;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto weights = NDArrayFactory::create<TypeParam>('c', {kH, kW, iC, oC});
    auto bias = NDArrayFactory::create<TypeParam>('c', {oC}, {1,2,3});
    auto gradO = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, oC});

    auto expGradI = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC},{ 0.226f,  0.343f,  0.46f,  0.577f, 1.172f,  1.46f,  1.748f,  2.036f, 1.892f,  2.288f,  2.684f,  3.08f, 1.284f,  1.581f,  1.878f,  2.175f, 4.458f,  5.133f,  5.808f,  6.483f, 6.186f,  7.023f,  7.86f,  8.697f,
                                                     3.39f,  3.93f,  4.47f,  5.01f, 9.642f, 10.803f, 11.964f, 13.125f,11.37f, 12.693f, 14.016f, 15.339f, 5.266f,  5.707f,  6.148f,  6.589f,12.98f, 13.916f, 14.852f, 15.788f,14.564f, 15.608f, 16.652f, 17.696f,
                                                     3.25f,  4.015f,  4.78f,  5.545f, 9.812f, 11.396f, 12.98f, 14.564f,10.532f, 12.224f, 13.916f, 15.608f, 9.708f, 10.977f, 12.246f, 13.515f,25.194f, 27.813f, 30.432f, 33.051f,26.922f, 29.703f, 32.484f, 35.265f,
                                                    11.814f, 13.326f, 14.838f, 16.35f,30.378f, 33.483f, 36.588f, 39.693f,32.106f, 35.373f, 38.64f, 41.907f,13.474f, 14.563f, 15.652f, 16.741f,31.988f, 34.22f, 36.452f, 38.684f,33.572f, 35.912f, 38.252f, 40.592f});

    auto expGradW = NDArrayFactory::create<TypeParam>('c', {kH, kW, iC, oC},{14.4f, 14.76f, 15.12f,14.4f, 14.76f, 15.12f,14.4f, 14.76f, 15.12f,14.4f, 14.76f, 15.12f, 9.24f,  9.48f,  9.72f, 9.24f,  9.48f,  9.72f, 9.24f,  9.48f,  9.72f, 9.24f,  9.48f,  9.72f,
                                                    17.04f, 17.52f, 18.f,17.04f, 17.52f, 18.f, 17.04f, 17.52f, 18.f, 17.04f, 17.52f, 18.f,10.88f, 11.2f, 11.52f,10.88f, 11.2f, 11.52f,10.88f, 11.2f, 11.52f,10.88f, 11.2f, 11.52f,
                                                    11.16f, 11.52f, 11.88f,11.16f, 11.52f, 11.88f,11.16f, 11.52f, 11.88f,11.16f, 11.52f, 11.88f, 7.08f,  7.32f,  7.56f, 7.08f,  7.32f,  7.56f, 7.08f,  7.32f,  7.56f, 7.08f,  7.32f,  7.56f});
    // auto expGradB('c', {oC},{});

    input = 2.;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.01, 0.01);

    nd4j::ops::conv2d_bp op;
    auto results = op.execute({&input, &weights, &bias, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto gradI = results->at(0);
    auto gradW = results->at(1);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    ASSERT_TRUE(expGradW.isSameShape(gradW));
    ASSERT_TRUE(expGradW.equalsTo(gradW));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_bp_test2) {

    int bS=2, iH=4,iW=3,  iC=4,oC=3,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto weights = NDArrayFactory::create<TypeParam>('c', {kH, kW, iC, oC});
    auto bias = NDArrayFactory::create<TypeParam>('c', {oC}, {1,2,3});
    auto gradO = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, oC});

    auto expGradI = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC},{ 0.014f, 0.032f, 0.05f, 0.068f,0.118f,0.181f, 0.244f, 0.307f,0.212f,0.257f, 0.302f, 0.347f,0.208f,0.298f, 0.388f, 0.478f,1.028f,1.262f, 1.496f, 1.73f,1.036f,1.18f, 1.324f, 1.468f,
                                                     0.928f,1.018f, 1.108f, 1.198f,2.9f,3.134f, 3.368f, 3.602f,2.188f,2.332f, 2.476f, 2.62f, 1.202f,1.274f, 1.346f, 1.418f,3.142f,3.313f, 3.484f, 3.655f,2.048f,2.147f, 2.246f, 2.345f,
                                                     0.086f,0.212f, 0.338f, 0.464f,0.694f,0.973f, 1.252f, 1.531f,0.716f,0.869f, 1.022f, 1.175f,1.216f,1.522f, 1.828f, 2.134f,3.908f,4.574f, 5.24f, 5.906f,2.908f,3.268f, 3.628f, 3.988f,
                                                     3.664f,3.97f, 4.276f, 4.582f,9.236f,9.902f,10.568f,11.234f,5.788f,6.148f, 6.508f, 6.868f,3.002f,3.182f, 3.362f, 3.542f,7.174f,7.561f, 7.948f, 8.335f,4.28f,4.487f, 4.694f, 4.901f});

    auto expGradW = NDArrayFactory::create<TypeParam>('c', {kH, kW, iC, oC},{1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,
                                                    1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,
                                                    1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f,1.84f, 2.f, 2.16f});
    // auto expGradB('c', {oC},{});

    input = 2.;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.01, 0.01);

    nd4j::ops::conv2d_bp op;
    auto results = op.execute({&input, &weights, &bias, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto gradI = results->at(0);
    auto gradW = results->at(1);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    ASSERT_TRUE(expGradW.isSameShape(gradW));
    ASSERT_TRUE(expGradW.equalsTo(gradW));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv2d_bp_test3) {
    int bS=2, iH=4,iW=3,  iC=4,oC=3,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    auto input = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto weights = NDArrayFactory::create<TypeParam>('c', {oC, iC, kH, kW});
    auto bias = NDArrayFactory::create<TypeParam>('c', {oC}, {1,2,3});
    auto gradO = NDArrayFactory::create<TypeParam>('c', {bS, oC, oH, oW});

    auto expGradI = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW},{ 0.567f, 1.224f, 0.66f, 1.314f, 2.82f, 1.512f, 1.386f, 2.976f, 1.596f, 0.801f, 1.71f, 0.912f, 0.657f, 1.422f, 0.768f, 1.53f, 3.288f, 1.764f, 1.602f, 3.444f, 1.848f, 0.927f, 1.98f, 1.056f, 
                                                     0.747f, 1.62f, 0.876f, 1.746f, 3.756f, 2.016f, 1.818f, 3.912f, 2.1f, 1.053f, 2.25f, 1.2f, 0.837f, 1.818f, 0.984f, 1.962f, 4.224f, 2.268f, 2.034f, 4.38f, 2.352f, 1.179f, 2.52f, 1.344f, 
                                                     1.467f, 3.06f, 1.596f, 3.186f, 6.636f, 3.456f, 3.402f, 7.08f, 3.684f, 1.845f, 3.834f, 1.992f, 1.773f, 3.69f, 1.92f, 3.834f, 7.968f, 4.14f, 4.05f, 8.412f, 4.368f, 2.187f, 4.536f, 2.352f, 
                                                     2.079f, 4.32f, 2.244f, 4.482f, 9.3f, 4.824f, 4.698f, 9.744f, 5.052f, 2.529f, 5.238f, 2.712f, 2.385f, 4.95f, 2.568f, 5.13f, 10.632f, 5.508f, 5.346f, 11.076f, 5.736f, 2.871f, 5.94f, 3.072f});

    auto expGradW = NDArrayFactory::create<TypeParam>('c', {oC, iC, kH, kW},{1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 
                                                    1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 1.3600e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 
                                                    2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.0000e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 
                                                    2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f, 2.6400e+00f});
    auto expGradB = NDArrayFactory::create<TypeParam>('c', {oC},{0.68f, 1.f, 1.32f});

    input = 2.;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.01, 0.01);
    weights.permutei({2,3,1,0});
    expGradW.permutei({2,3,1,0});

    nd4j::ops::conv2d_bp op;
    auto results = op.execute({&input, &weights, &bias, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto gradI = results->at(0);
    auto gradW = results->at(1);
    auto gradB = results->at(2);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    ASSERT_TRUE(expGradW.isSameShape(gradW));
    ASSERT_TRUE(expGradW.equalsTo(gradW));

    ASSERT_TRUE(expGradB.isSameShape(gradB));
    ASSERT_TRUE(expGradB.equalsTo(gradB));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, conv2d_bp_4) {

    int bS=1, iH=7,iW=1,  iC=2,oC=3,  kH=2,kW=1,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=7,oW=1;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iC, iH, iW}, nd4j::DataType::FLOAT32);
    NDArray weights('c', {kH, kW, iC, oC}, nd4j::DataType::FLOAT32);
    NDArray bias('c', {oC}, {1,2,3}, nd4j::DataType::FLOAT32);
    NDArray gradO('c', {bS, oC, oH, oW}, nd4j::DataType::FLOAT32);

    NDArray gradI('c', {bS, iC, iH, iW}, nd4j::DataType::FLOAT32);
    NDArray gradW('c', {kH, kW, iC, oC}, nd4j::DataType::FLOAT32);
    NDArray gradB('c', {oC}, nd4j::DataType::FLOAT32);


    input = 2.;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.01, 0.01);

    nd4j::ops::conv2d_bp op;
    auto status = op.execute({&input, &weights, &bias, &gradO}, {&gradI, &gradW, &gradB}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat}, {});

    ASSERT_EQ(Status::OK(), status);
}

////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_bp_test1) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int       oD=3,oH=4,oW=3;
    int paddingMode = 1;             // 1-SAME,  0-VALID;
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});
    auto bias     = NDArrayFactory::create<TypeParam>('c', {oC}, {1,2,3});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, oC});

    auto expGradI = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC},{0.226f, 0.343f, 0.46f, 0.577f, 1.172f, 1.46f, 1.748f, 2.036f, 1.892f, 2.288f, 2.684f, 3.08f, 1.284f, 1.581f, 1.878f, 2.175f, 4.458f, 5.133f, 5.808f, 6.483f, 6.186f, 7.023f, 7.86f, 8.697f, 3.39f, 3.93f, 4.47f, 5.01f, 9.642f, 10.803f, 11.964f, 13.125f, 11.37f, 12.693f, 14.016f, 15.339f, 
                                                        5.266f, 5.707f, 6.148f, 6.589f, 12.98f, 13.916f, 14.852f, 15.788f, 14.564f, 15.608f, 16.652f, 17.696f, 6.284f, 7.166f, 8.048f, 8.93f, 17.896f, 19.768f, 21.64f, 23.512f, 21.928f, 24.016f, 26.104f, 28.192f, 18.12f, 19.686f, 21.252f, 22.818f, 45.852f, 49.146f, 52.44f, 55.734f, 53.196f, 56.814f, 60.432f, 64.05f, 
                                                       28.164f, 30.216f, 32.268f, 34.32f, 67.884f, 72.15f, 76.416f, 80.682f, 75.228f, 79.818f, 84.408f, 88.998f, 29.324f, 30.854f, 32.384f, 33.914f, 67.432f, 70.6f, 73.768f, 76.936f, 73.192f, 76.576f, 79.96f, 83.344f, 27.884f, 30.062f, 32.24f, 34.418f, 66.28f, 70.744f, 75.208f, 79.672f, 70.312f, 74.992f, 79.672f, 84.352f, 
                                                       58.296f, 61.806f, 65.316f, 68.826f, 133.98f, 141.162f, 148.344f, 155.526f, 141.324f, 148.83f, 156.336f, 163.842f, 68.34f, 72.336f, 76.332f, 80.328f, 156.012f, 164.166f, 172.32f, 180.474f, 163.356f, 171.834f, 180.312f, 188.79f, 61.292f, 64.118f, 66.944f, 69.77f, 136.552f, 142.312f, 148.072f, 153.832f, 142.312f, 148.288f, 154.264f, 160.24f, 
                                                        9.298f, 11.359f, 13.42f, 15.481f, 27.092f, 31.268f, 35.444f, 39.62f, 27.812f, 32.096f, 36.38f, 40.664f, 26.556f, 29.769f, 32.982f, 36.195f, 66.666f, 73.173f, 79.68f, 86.187f, 68.394f, 75.063f, 81.732f, 88.401f, 28.662f, 32.118f, 35.574f, 39.03f, 71.85f, 78.843f, 85.836f, 92.829f, 73.578f, 80.733f, 87.888f, 95.043f, 
                                                       29.89f, 32.275f, 34.66f, 37.045f, 70.004f, 74.828f, 79.652f, 84.476f, 71.588f, 76.52f, 81.452f, 86.384f, 71.084f, 75.854f, 80.624f, 85.394f, 163.048f, 172.696f, 182.344f, 191.992f, 167.08f, 176.944f, 186.808f, 196.672f, 138.648f, 146.046f, 153.444f, 160.842f, 310.236f, 325.194f, 340.152f, 355.11f, 317.58f, 332.862f, 348.144f, 363.426f, 
                                                      148.692f, 156.576f, 164.46f, 172.344f, 332.268f, 348.198f, 364.128f, 380.058f, 339.612f, 355.866f, 372.12f, 388.374f, 125.228f, 130.646f, 136.064f, 141.482f, 274.792f, 285.736f, 296.68f, 307.624f, 280.552f, 291.712f, 302.872f, 314.032f, 92.684f, 98.75f, 104.816f, 110.882f, 211.432f, 223.672f, 235.912f, 248.152f, 215.464f, 227.92f, 240.376f, 252.832f, 
                                                      178.824f, 188.166f, 197.508f, 206.85f, 398.364f, 417.21f, 436.056f, 454.902f, 405.708f, 424.878f, 444.048f, 463.218f, 188.868f, 198.696f, 208.524f, 218.352f, 420.396f, 440.214f, 460.032f, 479.85f, 427.74f, 447.882f, 468.024f, 488.166f, 157.196f, 163.91f, 170.624f, 177.338f, 343.912f, 357.448f, 370.984f, 384.52f, 349.672f, 363.424f, 377.176f, 390.928f});

    auto expGradW = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC},{120.96f, 122.04f, 123.12f, 120.96f, 122.04f, 123.12f, 120.96f, 122.04f, 123.12f, 120.96f, 122.04f, 123.12f, 79.56f, 80.28f, 81.f, 79.56f, 80.28f, 81.f, 79.56f, 80.28f, 81.f, 79.56f, 80.28f, 81.f, 
                                                        154.8f, 156.24f, 157.68f, 154.8f, 156.24f, 157.68f, 154.8f, 156.24f, 157.68f, 154.8f, 156.24f, 157.68f, 101.76f, 102.72f, 103.68f, 101.76f, 102.72f, 103.68f, 101.76f, 102.72f, 103.68f, 101.76f, 102.72f, 103.68f, 
                                                        111.24f, 112.32f, 113.4f, 111.24f, 112.32f, 113.4f, 111.24f, 112.32f, 113.4f, 111.24f, 112.32f, 113.4f, 73.08f, 73.8f, 74.52f, 73.08f, 73.8f, 74.52f, 73.08f, 73.8f, 74.52f, 73.08f, 73.8f, 74.52f, 
                                                         67.68f, 68.4f, 69.12f, 67.68f, 68.4f, 69.12f, 67.68f, 68.4f, 69.12f, 67.68f, 68.4f, 69.12f, 44.4f, 44.88f, 45.36f, 44.4f, 44.88f, 45.36f, 44.4f, 44.88f, 45.36f, 44.4f, 44.88f, 45.36f, 
                                                         85.92f, 86.88f, 87.84f, 85.92f, 86.88f, 87.84f, 85.92f, 86.88f, 87.84f, 85.92f, 86.88f, 87.84f, 56.32f, 56.96f, 57.6f, 56.32f, 56.96f, 57.6f, 56.32f, 56.96f, 57.6f, 56.32f, 56.96f, 57.6f, 
                                                         61.2f, 61.92f, 62.64f, 61.2f, 61.92f, 62.64f, 61.2f, 61.92f, 62.64f, 61.2f, 61.92f, 62.64f, 40.08f, 40.56f, 41.04f, 40.08f, 40.56f, 41.04f, 40.08f, 40.56f, 41.04f, 40.08f, 40.56f, 41.04f});
    // auto expGradB('c', {oC},{});

    input = 2.;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.01, 0.01);

    nd4j::ops::conv3dnew_bp op;
    auto results = op.execute({&input, &weights, &bias, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto gradI = results->at(0);
    auto gradW = results->at(1);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    ASSERT_TRUE(expGradW.isSameShape(gradW));
    ASSERT_TRUE(expGradW.equalsTo(gradW));

    delete results;
}


////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_bp_test2) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int       oD=2,oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME,  0-VALID;
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});
    auto bias     = NDArrayFactory::create<TypeParam>('c', {oC}, {1,2,3});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, oC});

    auto expGradI = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC},{ 0.014f, 0.032f, 0.05f, 0.068f, 0.118f, 0.181f, 0.244f, 0.307f, 0.212f, 0.257f, 0.302f, 0.347f, 0.208f, 0.298f, 0.388f, 0.478f, 1.028f, 1.262f, 1.496f, 1.73f, 1.036f, 1.18f, 1.324f, 1.468f, 0.928f, 1.018f, 1.108f, 1.198f, 2.9f, 3.134f, 3.368f, 3.602f, 2.188f, 2.332f, 2.476f, 2.62f, 
                                                         1.202f, 1.274f, 1.346f, 1.418f, 3.142f, 3.313f, 3.484f, 3.655f, 2.048f, 2.147f, 2.246f, 2.345f, 0.532f, 0.676f, 0.82f, 0.964f, 2.324f, 2.666f, 3.008f, 3.35f, 2.008f, 2.206f, 2.404f, 2.602f, 3.584f, 3.98f, 4.376f, 4.772f, 10.552f, 11.452f, 12.352f, 13.252f, 7.4f, 7.904f, 8.408f, 8.912f, 
                                                         6.752f, 7.148f, 7.544f, 7.94f, 17.752f, 18.652f, 19.552f, 20.452f, 11.432f, 11.936f, 12.44f, 12.944f, 5.932f, 6.184f, 6.436f, 6.688f, 14.42f, 14.978f, 15.536f, 16.094f, 8.704f, 9.01f, 9.316f, 9.622f, 3.11f, 3.236f, 3.362f, 3.488f, 7.39f, 7.669f, 7.948f, 8.227f, 4.388f, 4.541f, 4.694f, 4.847f, 
                                                         8.56f, 8.866f, 9.172f, 9.478f, 19.892f, 20.558f, 21.224f, 21.89f, 11.548f, 11.908f, 12.268f, 12.628f, 11.008f, 11.314f, 11.62f, 11.926f, 25.22f, 25.886f, 26.552f, 27.218f, 14.428f, 14.788f, 15.148f, 15.508f, 7.322f, 7.502f, 7.682f, 7.862f, 16.462f, 16.849f, 17.236f, 17.623f, 9.248f, 9.455f, 9.662f, 9.869f, 
                                                         0.158f, 0.392f, 0.626f, 0.86f, 1.27f, 1.765f, 2.26f, 2.755f, 1.22f, 1.481f, 1.742f, 2.003f, 2.224f, 2.746f, 3.268f, 3.79f, 6.788f, 7.886f, 8.984f, 10.082f, 4.78f, 5.356f, 5.932f, 6.508f, 6.4f, 6.922f, 7.444f, 7.966f, 15.572f, 16.67f, 17.768f, 18.866f, 9.388f, 9.964f, 10.54f, 11.116f, 
                                                         4.802f, 5.09f, 5.378f, 5.666f, 11.206f, 11.809f, 12.412f, 13.015f, 6.512f, 6.827f, 7.142f, 7.457f, 6.004f, 6.58f, 7.156f, 7.732f, 14.996f, 16.202f, 17.408f, 18.614f, 9.208f, 9.838f, 10.468f, 11.098f, 17.984f, 19.244f, 20.504f, 21.764f, 42.808f, 45.436f, 48.064f, 50.692f, 25.256f, 26.624f, 27.992f, 29.36f, 
                                                        28.064f, 29.324f, 30.584f, 31.844f, 63.832f, 66.46f, 69.088f, 71.716f, 36.2f, 37.568f, 38.936f, 40.304f, 18.316f, 19.f, 19.684f, 20.368f, 40.916f, 42.338f, 43.76f, 45.182f, 22.816f, 23.554f, 24.292f, 25.03f, 8.438f, 8.78f, 9.122f, 9.464f, 18.91f, 19.621f, 20.332f, 21.043f, 10.58f, 10.949f, 11.318f, 11.687f, 
                                                        20.944f, 21.682f, 22.42f, 23.158f, 46.388f, 47.918f, 49.448f, 50.978f, 25.66f, 26.452f, 27.244f, 28.036f, 26.848f, 27.586f, 28.324f, 29.062f, 58.628f, 60.158f, 61.688f, 63.218f, 31.996f, 32.788f, 33.58f, 34.372f, 16.106f, 16.502f, 16.898f, 17.294f, 34.894f, 35.713f, 36.532f, 37.351f, 18.896f, 19.319f, 19.742f, 20.165f});

    auto expGradW = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC},{7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 
                                                        7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 
                                                        7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 
                                                        7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f, 7.52f, 7.84f, 8.16f});
    // auto expGradB('c', {oC},{});

    input = 2.;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.01, 0.01);

    nd4j::ops::conv3dnew_bp op;
    auto results = op.execute({&input, &weights, &bias, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto gradI = results->at(0);
    auto gradW = results->at(1);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    ASSERT_TRUE(expGradW.isSameShape(gradW));
    ASSERT_TRUE(expGradW.equalsTo(gradW));

    delete results;
}


////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_bp_test3) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int       oD=2,oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME,  0-VALID;
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {oC, iC, kD, kH, kW});
    auto bias     = NDArrayFactory::create<TypeParam>('c', {oC}, {1,2,3});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oC, oD, oH, oW});

    auto expGradI = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW},{2.091f, 4.356f, 2.268f, 4.53f, 9.42f, 4.896f, 4.65f, 9.672f, 5.028f, 2.517f, 5.226f, 2.712f, 4.932f, 10.242f, 5.316f, 10.62f, 22.02f, 11.412f, 10.908f, 22.62f, 11.724f, 5.868f, 12.15f, 6.288f, 2.913f, 6.03f, 3.12f, 6.234f, 12.888f, 6.66f, 6.402f, 13.236f, 6.84f, 3.423f, 7.068f, 3.648f, 
                                                        2.415f, 5.04f, 2.628f, 5.25f, 10.932f, 5.688f, 5.37f, 11.184f, 5.82f, 2.913f, 6.054f, 3.144f, 5.724f, 11.898f, 6.18f, 12.348f, 25.62f, 13.284f, 12.636f, 26.22f, 13.596f, 6.804f, 14.094f, 7.296f, 3.381f, 7.002f, 3.624f, 7.242f, 14.976f, 7.74f, 7.41f, 15.324f, 7.92f, 3.963f, 8.184f, 4.224f, 
                                                        2.739f, 5.724f, 2.988f, 5.97f, 12.444f, 6.48f, 6.09f, 12.696f, 6.612f, 3.309f, 6.882f, 3.576f, 6.516f, 13.554f, 7.044f, 14.076f, 29.22f, 15.156f, 14.364f, 29.82f, 15.468f, 7.74f, 16.038f, 8.304f, 3.849f, 7.974f, 4.128f, 8.25f, 17.064f, 8.82f, 8.418f, 17.412f, 9.f, 4.503f, 9.3f, 4.8f, 
                                                        3.063f, 6.408f, 3.348f, 6.69f, 13.956f, 7.272f, 6.81f, 14.208f, 7.404f, 3.705f, 7.71f, 4.008f, 7.308f, 15.21f, 7.908f, 15.804f, 32.82f, 17.028f, 16.092f, 33.42f, 17.34f, 8.676f, 17.982f, 9.312f, 4.317f, 8.946f, 4.632f, 9.258f, 19.152f, 9.9f, 9.426f, 19.5f, 10.08f, 5.043f, 10.416f, 5.376f, 
                                                        5.619f, 11.484f, 5.868f, 11.73f, 23.964f, 12.24f, 12.138f, 24.792f, 12.66f, 6.333f, 12.93f, 6.6f, 12.42f, 25.362f, 12.948f, 25.884f, 52.836f, 26.964f, 26.748f, 54.588f, 27.852f, 13.932f, 28.422f, 14.496f, 6.873f, 14.022f, 7.152f, 14.298f, 29.16f, 14.868f, 14.754f, 30.084f, 15.336f, 7.671f, 15.636f, 7.968f, 
                                                        6.807f, 13.896f, 7.092f, 14.178f, 28.932f, 14.76f, 14.586f, 29.76f, 15.18f, 7.593f, 15.486f, 7.896f, 14.94f, 30.474f, 15.54f, 31.068f, 63.348f, 32.292f, 31.932f, 65.1f, 33.18f, 16.596f, 33.822f, 17.232f, 8.205f, 16.722f, 8.52f, 17.034f, 34.704f, 17.676f, 17.49f, 35.628f, 18.144f, 9.075f, 18.48f, 9.408f, 
                                                        7.995f, 16.308f, 8.316f, 16.626f, 33.9f, 17.28f, 17.034f, 34.728f, 17.7f, 8.853f, 18.042f, 9.192f, 17.46f, 35.586f, 18.132f, 36.252f, 73.86f, 37.62f, 37.116f, 75.612f, 38.508f, 19.26f, 39.222f, 19.968f, 9.537f, 19.422f, 9.888f, 19.77f, 40.248f, 20.484f, 20.226f, 41.172f, 20.952f, 10.479f, 21.324f, 10.848f, 
                                                        9.183f, 18.72f, 9.54f, 19.074f, 38.868f, 19.8f, 19.482f, 39.696f, 20.22f, 10.113f, 20.598f, 10.488f, 19.98f, 40.698f, 20.724f, 41.436f, 84.372f, 42.948f, 42.3f, 86.124f, 43.836f, 21.924f, 44.622f, 22.704f, 10.869f, 22.122f, 11.256f, 22.506f, 45.792f, 23.292f, 22.962f, 46.716f, 23.76f, 11.883f, 24.168f, 12.288f});

    auto expGradW = NDArrayFactory::create<TypeParam>('c', {oC, iC, kD, kH, kW},{5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 
                                                        5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 5.28f, 
                                                        7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 
                                                        7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 7.84f, 
                                                        10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 
                                                        10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f, 10.4f});

    auto expGradB = NDArrayFactory::create<TypeParam>('c', {oC},{2.64f, 3.92f, 5.2f});

    input = 2.;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.01, 0.01);
    weights.permutei({2, 3, 4, 1, 0});
    expGradW.permutei({2, 3, 4, 1, 0});

    nd4j::ops::conv3dnew_bp op;
    auto results = op.execute({&input, &weights, &bias, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* gradI = results->at(0);
    auto* gradW = results->at(1);
    auto* gradB = results->at(2);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    ASSERT_TRUE(expGradW.isSameShape(gradW));
    ASSERT_TRUE(expGradW.equalsTo(gradW));

    ASSERT_TRUE(expGradB.isSameShape(gradB));
    ASSERT_TRUE(expGradB.equalsTo(gradB));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, depthwise_conv2d_1) {

    int bS=2, iH=4,iW=3,  iC=2,mC=2,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oC=iC*mC;
    int       oH=4,oW=3;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kH, kW, iC, mC});


    auto expOutput = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, oC},{12.f, 12.8f, 13.6f, 14.4f, 12.f, 12.8f, 13.6f, 14.4f, 5.2f, 5.6f, 6.f, 6.4f, 13.2f, 14.4f, 15.6f, 16.8f, 13.2f, 14.4f, 15.6f, 16.8f, 5.4f, 6.f, 6.6f, 7.2f, 
                                                     13.2f, 14.4f, 15.6f, 16.8f, 13.2f, 14.4f, 15.6f, 16.8f, 5.4f, 6.f, 6.6f, 7.2f, 5.6f, 6.4f, 7.2f, 8.f, 5.6f, 6.4f, 7.2f, 8.f, 2.f, 2.4f, 2.8f, 3.2f, 
                                                     12.f, 12.8f, 13.6f, 14.4f, 12.f, 12.8f, 13.6f, 14.4f, 5.2f, 5.6f, 6.f, 6.4f, 13.2f, 14.4f, 15.6f, 16.8f, 13.2f, 14.4f, 15.6f, 16.8f, 5.4f, 6.f, 6.6f, 7.2f, 
                                                     13.2f, 14.4f, 15.6f, 16.8f, 13.2f, 14.4f, 15.6f, 16.8f, 5.4f, 6.f, 6.6f, 7.2f, 5.6f, 6.4f, 7.2f, 8.f, 5.6f, 6.4f, 7.2f, 8.f, 2.f, 2.4f, 2.8f, 3.2f});
    input = 2.;
    weights.linspace(0.1, 0.1);

    nd4j::ops::depthwise_conv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, depthwise_conv2d_2) {

    int bS=2, iH=4,iW=3,  iC=2,mC=2,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oC=iC*mC;
    int       oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<double>('c', {bS, iH, iW, iC});
    auto weights  = NDArrayFactory::create<double>('c', {kH, kW, iC, mC});


    auto expOutput = NDArrayFactory::create<double>('c', {bS, oH, oW, oC},{13.2f,  14.4f,  15.6f,  16.8f,  13.2f,  14.4f,  15.6f,  16.8f,  13.2f,  14.4f,  15.6f,  16.8f,  13.2f,  14.4f,  15.6f,  16.8f, 
                                                     13.2f,  14.4f,  15.6f,  16.8f, 13.2f,  14.4f,  15.6f,  16.8f,  13.2f,  14.4f,  15.6f,  16.8f, 13.2f,  14.4f,  15.6f,  16.8f});
    input = 2.;
    weights.linspace(0.1, 0.1);

    nd4j::ops::depthwise_conv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}


//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, depthwise_conv2d_3) {

    int bS=2, iH=4,iW=3,  iC=2,mC=2,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oC=iC*mC;
    int       oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<double>('c', {bS, iC, iH, iW});
    auto weights  = NDArrayFactory::create<double>('c', {mC, iC, kH, kW});
    auto biases   = NDArrayFactory::create<double>('c', {iC*mC}, {1,2,3,4});


    auto expOutput = NDArrayFactory::create<double>('c', {bS, oC, oH, oW},{5.2, 5.2, 5.2, 5.2,20.6,20.6,20.6,20.6,14.4,14.4,14.4,14.4,29.8,29.8,29.8,29.8, 5.2, 5.2, 5.2, 5.2,20.6,20.6,20.6,20.6,14.4,14.4,14.4,14.4,29.8,29.8,29.8,29.8});
    input = 2.;
    weights.linspace(0.1, 0.1);
    weights.permutei({2,3,1,0});

    nd4j::ops::depthwise_conv2d op;
    auto results = op.execute({&input, &weights, &biases}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    // output->printIndexedBuffer();

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, depthwise_conv2d_4) {

    int bS=1, iH=111,iW=111,  iC=32,mC=1,  kH=7,kW=7,  sH=2,sW=2,  pH=0,pW=0,  dH=1,dW=1;
    int       oC=iC*mC;
    int       oH=56,oW=56;

    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    const float unique = -1000000;

    NDArray input('c', {bS, iH, iW, iC}, nd4j::DataType::FLOAT32);
    NDArray weights('c', {kH, kW, iC, mC}, nd4j::DataType::FLOAT32);
    NDArray output('c', {bS, oH, oW, oC}, nd4j::DataType::FLOAT32);
    input.linspace(0.1, 0.0001);
    weights = 0.5;
    output = unique;

    nd4j::ops::depthwise_conv2d op;
    Nd4jStatus status = op.execute({&input, &weights}, {&output} , {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat}, {});

    ASSERT_EQ(Status::OK(), status);

    for(Nd4jLong i=output.lengthOf()/1.5; i < output.lengthOf(); ++i)
        ASSERT_EQ(output.e<float>(i) != unique, true);
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, depthwise_conv2d_5) {

    int bS=1, iH=3,iW=3,  iC=2,mC=1,  kH=2,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oC=iC*mC;
    int       oH=3,oW=3;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input   = NDArrayFactory::create<double>('c', {bS, iH, iW, iC});
    auto weights = NDArrayFactory::create<double>('c', {kH, kW, iC, mC});


    auto expOutput = NDArrayFactory::create<double>('c', {bS, oH, oW, oC}, {20., 24.,28., 32.,16., 18.,44., 48.,52., 56.,28., 30.,28., 30.,32., 34.,17., 18.});
    input.linspace(1.);
    weights = 1.;

    nd4j::ops::depthwise_conv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto output = results->at(0);
    // output->printIndexedBuffer();

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, depthwise_conv2d_6) {

    int bS=1, iH=3,iW=3,  iC=2,mC=1,  kH=2,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oC=iC*mC;
    int       oH=3,oW=3;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iH, iW, iC}, nd4j::DataType::DOUBLE);
    NDArray weights('c', {kH, kW, iC, mC}, nd4j::DataType::DOUBLE);

    NDArray expOutput('c', {bS, oH, oW, oC}, {20., 24.,28., 32.,16., 18.,44., 48.,52., 56.,28., 30.,28., 30.,32., 34.,17., 18.});
    input.linspace(1.);
    weights = 1.;

    nd4j::ops::depthwise_conv2d op;
    ResultSet* results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    NDArray* output = results->at(0);
    // output.printIndexedBuffer();

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, depthwise_conv2d_7) {

    int bS=1, iH=3,iW=3,  iC=2,mC=2,  kH=1,kW=1,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oC=iC*mC;
    int       oH=3,oW=3;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iC, iH, iW}, {0.6793503761291504, 0.35508695244789124, 0.842789351940155, 0.20031332969665527, 0.7014986872673035, 0.3106933832168579,
                                        0.44793984293937683, 0.9380097389221191, 0.3266739547252655, 0.15187257528305054, 0.3833175301551819, 0.7821229696273804,
                                        0.19880719482898712, 0.7985635995864868, 0.16326339542865753, 0.14696824550628662, 0.2608966827392578, 0.13505761325359344});
    NDArray weights('c', {kH, kW, iC, mC}, {0.1308445781469345, 0.6442840099334717, 0.5698848366737366, 0.19896849989891052});
    NDArray biases('c', {1,iC*mC}, {0.6123566627502441, 0.37637925148010254, 0.17464971542358398, 0.4270855486392975});

    NDArray expOutput('c', {bS, oC, oH, oW}, {0.7012459761288241, 0.6588178652487691, 0.722631079971582, 0.6385665758716108, 0.7041439625563628, 0.6530092074102978,
        0.670967162534851, 0.735090151337225, 0.6551001785478623, 0.8140738359624038, 0.6051560970782859, 0.9193749546773375, 0.5054379267801892, 0.8283436386757472,
        0.5765540302788565, 0.6649797296980537, 0.9807239274294943, 0.586850056971322, 0.261199593183985, 0.3930965634902499, 0.6203697362284615, 0.28794692117826504,
        0.6297390019475202, 0.26769104886224415, 0.25840469001015975, 0.3233307788551656, 0.25161700129415276, 0.4573034071191504, 0.5033536625992294, 0.5827033826425385,
        0.4666419179635315, 0.585974550122895, 0.4595698215161401, 0.45632759998045813, 0.4789957702325296, 0.4539577593482922});


    nd4j::ops::depthwise_conv2d op;
    auto results = op.execute({&input, &weights, &biases}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, depthwise_conv2d_bp_test1) {

    int bS=2, iH=4,iW=3,  iC=2,mC=2,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=4,oW=3;
    int       oC=iC*mC;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<double>('c', {bS, iH, iW, iC});
    auto weights  = NDArrayFactory::create<double>('c', {kH, kW, iC, mC});
    auto bias     = NDArrayFactory::create<double>('c', {oC}, {1,2,3,4});
    auto gradO    = NDArrayFactory::create<double>('c', {bS, oH, oW, oC});

    auto expGradI = NDArrayFactory::create<double>('c', {bS, iH, iW, iC},{0.07 ,  0.19 , 0.348,  0.652, 0.588,  0.956, 0.387,  0.687, 1.326,  2.022, 1.878,  2.67 , 1.071,  1.515, 2.982,  3.966, 3.534,  4.614, 1.606,  1.982, 3.932,  4.748, 4.428,  5.308,
                                                    1.126,  1.63 , 3.228,  4.3  , 3.468,  4.604, 3.123,  3.999, 7.95 ,  9.798, 8.502, 10.446, 3.807,  4.827, 9.606, 11.742,10.158, 12.39 , 4.198,  4.958, 9.884, 11.468,10.38 , 12.028});

    auto expGradW = NDArrayFactory::create<double>('c', {kH, kW, iC, mC},{19.08, 19.44,19.8 , 20.16,12.24, 12.48,12.72, 12.96,22.56, 23.04,23.52, 24. ,14.4 , 14.72,15.04, 15.36,14.76, 15.12,15.48, 15.84, 9.36,  9.6 , 9.84, 10.08});

    input = 2.;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.01, 0.01);

    nd4j::ops::depthwise_conv2d_bp op;
    auto results = op.execute({&input, &weights, &bias, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto* gradI = results->at(0);
    auto* gradW = results->at(1);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    ASSERT_TRUE(expGradW.isSameShape(gradW));
    ASSERT_TRUE(expGradW.equalsTo(gradW));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, depthwise_conv2d_bp_test2) {

    int bS=2, iH=4,iW=3,  iC=2,mC=2,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=2,oW=2;
    int       oC=iC*mC;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<double>('c', {bS, iH, iW, iC});
    auto weights  = NDArrayFactory::create<double>('c', {kH, kW, iC, mC});
    auto bias     = NDArrayFactory::create<double>('c', {oC}, {1,2,3,4});
    auto gradO    = NDArrayFactory::create<double>('c', {bS, oH, oW, oC});

    auto expGradI = NDArrayFactory::create<double>('c', {bS, iH, iW, iC},{0.005, 0.025,0.034, 0.106,0.061, 0.113,0.058, 0.162,0.292, 0.564,0.298, 0.466,0.234, 0.402,0.772, 1.172,0.602, 0.834,0.333, 0.449,0.882, 1.146,0.581, 0.729,
                                                    0.053, 0.137,0.258, 0.458,0.237, 0.353,0.41 , 0.642,1.252, 1.78 ,0.906, 1.202,1.098, 1.394,2.756, 3.412,1.722, 2.082,0.893, 1.073,2.13 , 2.522,1.269, 1.481});
    auto expGradW = NDArrayFactory::create<double>('c', {kH, kW, iC, mC},{2.4 , 2.56,2.72, 2.88,2.4 , 2.56,2.72, 2.88,2.4 , 2.56,2.72, 2.88,2.4 , 2.56,2.72, 2.88,2.4 , 2.56,2.72, 2.88,2.4 , 2.56,2.72, 2.88});

    input = 2.;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.01, 0.01);

    nd4j::ops::depthwise_conv2d_bp op;
    auto results = op.execute({&input, &weights, &bias, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto* gradI = results->at(0);
    auto* gradW = results->at(1);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    ASSERT_TRUE(expGradW.isSameShape(gradW));
    ASSERT_TRUE(expGradW.equalsTo(gradW));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test1) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int paddingMode = 1;             // 1-SAME,  0-VALID;
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {2, 3, 4, 3, 3}, {64.f, 64.f, 64.f, 64.f, 64.f, 64.f, 32.f, 32.f, 32.f, 96.f, 96.f, 96.f, 96.f, 96.f, 96.f, 48.f, 48.f, 48.f, 96.f, 96.f, 96.f, 96.f, 96.f, 96.f, 48.f, 48.f, 48.f, 
                                                   64.f, 64.f, 64.f, 64.f, 64.f, 64.f, 32.f, 32.f, 32.f, 64.f, 64.f, 64.f, 64.f, 64.f, 64.f, 32.f, 32.f, 32.f, 96.f, 96.f, 96.f, 96.f, 96.f, 96.f, 48.f, 48.f, 48.f, 
                                                   96.f, 96.f, 96.f, 96.f, 96.f, 96.f, 48.f, 48.f, 48.f, 64.f, 64.f, 64.f, 64.f, 64.f, 64.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 16.f, 16.f, 16.f, 
                                                   48.f, 48.f, 48.f, 48.f, 48.f, 48.f, 24.f, 24.f, 24.f, 48.f, 48.f, 48.f, 48.f, 48.f, 48.f, 24.f, 24.f, 24.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 16.f, 16.f, 16.f, 
                                                   64.f, 64.f, 64.f, 64.f, 64.f, 64.f, 32.f, 32.f, 32.f, 96.f, 96.f, 96.f, 96.f, 96.f, 96.f, 48.f, 48.f, 48.f, 96.f, 96.f, 96.f, 96.f, 96.f, 96.f, 48.f, 48.f, 48.f, 
                                                   64.f, 64.f, 64.f, 64.f, 64.f, 64.f, 32.f, 32.f, 32.f, 64.f, 64.f, 64.f, 64.f, 64.f, 64.f, 32.f, 32.f, 32.f, 96.f, 96.f, 96.f, 96.f, 96.f, 96.f, 48.f, 48.f, 48.f, 
                                                   96.f, 96.f, 96.f, 96.f, 96.f, 96.f, 48.f, 48.f, 48.f, 64.f, 64.f, 64.f, 64.f, 64.f, 64.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 16.f, 16.f, 16.f, 
                                                   48.f, 48.f, 48.f, 48.f, 48.f, 48.f, 24.f, 24.f, 24.f, 48.f, 48.f, 48.f, 48.f, 48.f, 48.f, 24.f, 24.f, 24.f, 32.f, 32.f, 32.f, 32.f, 32.f, 32.f, 16.f, 16.f, 16.f});
    input = 2.;
    weights = 1.;

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expected.isSameShape(output));
    ASSERT_TRUE(expected.equalsTo(output));

    delete results;
}


//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test2) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int paddingMode = 1;             // 1-SAME,  0-VALID;
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {2, 3, 4, 3, 3}, {534.4f, 540.8f, 547.2f, 534.4f, 540.8f, 547.2f, 248.f, 251.2f, 254.4f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 314.4f, 319.2f, 324.f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 314.4f, 319.2f, 324.f, 
                                                   380.8f, 387.2f, 393.6f, 380.8f, 387.2f, 393.6f, 171.2f, 174.4f, 177.6f, 534.4f, 540.8f, 547.2f, 534.4f, 540.8f, 547.2f, 248.f, 251.2f, 254.4f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 314.4f, 319.2f, 324.f, 
                                                   686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 314.4f, 319.2f, 324.f, 380.8f, 387.2f, 393.6f, 380.8f, 387.2f, 393.6f, 171.2f, 174.4f, 177.6f, 152.f, 155.2f, 158.4f, 152.f, 155.2f, 158.4f, 66.4f, 68.f, 69.6f, 
                                                   170.4f, 175.2f, 180.f, 170.4f, 175.2f, 180.f, 70.8f, 73.2f, 75.6f, 170.4f, 175.2f, 180.f, 170.4f, 175.2f, 180.f, 70.8f, 73.2f, 75.6f, 75.2f, 78.4f, 81.6f, 75.2f, 78.4f, 81.6f, 28.f, 29.6f, 31.2f, 
                                                   534.4f, 540.8f, 547.2f, 534.4f, 540.8f, 547.2f, 248.f, 251.2f, 254.4f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 314.4f, 319.2f, 324.f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 314.4f, 319.2f, 324.f, 
                                                   380.8f, 387.2f, 393.6f, 380.8f, 387.2f, 393.6f, 171.2f, 174.4f, 177.6f, 534.4f, 540.8f, 547.2f, 534.4f, 540.8f, 547.2f, 248.f, 251.2f, 254.4f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 314.4f, 319.2f, 324.f, 
                                                   686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 314.4f, 319.2f, 324.f, 380.8f, 387.2f, 393.6f, 380.8f, 387.2f, 393.6f, 171.2f, 174.4f, 177.6f, 152.f, 155.2f, 158.4f, 152.f, 155.2f, 158.4f, 66.4f, 68.f, 69.6f, 
                                                   170.4f, 175.2f, 180.f, 170.4f, 175.2f, 180.f, 70.8f, 73.2f, 75.6f, 170.4f, 175.2f, 180.f, 170.4f, 175.2f, 180.f, 70.8f, 73.2f, 75.6f, 75.2f, 78.4f, 81.6f, 75.2f, 78.4f, 81.6f, 28.f, 29.6f, 31.2f});
    input = 2.;
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expected.isSameShape(output));
    ASSERT_TRUE(expected.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test3) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int paddingMode = 0;             // 1-SAME,  0-VALID;
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {2, 2, 2, 2, 3},  {686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 
                                                    686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 
                                                    686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 
                                                    686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f, 686.4f, 696.f, 705.6f});
    input = 2.;
    weights.linspace(0.1, 0.1);

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expected.isSameShape(output));
    ASSERT_TRUE(expected.equalsTo(output));

    delete results;
}


//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test4) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int paddingMode = 0;             // 1-SAME,  0-VALID;
    int dataFormat =  0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {2, 3, 2, 2, 2});
    input = 2.;
    weights = 0.5;
    expected = 48.;

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expected.isSameShape(output));
    ASSERT_TRUE(expected.equalsTo(output));

    delete results;
}

////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test5) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int paddingMode = 0;             // 1-SAME,  0-VALID;
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});
    auto bias     = NDArrayFactory::create<TypeParam>('c', {oC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {2, 3, 2, 2, 2});

    input = 2.;
    weights = 0.5;
    expected = 49.;
    bias = 1.;

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    // output->printIndexedBuffer();

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expected.isSameShape(output));
    ASSERT_TRUE(expected.equalsTo(output));

    delete results;
}

////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test6) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int paddingMode = 0;             // 1-SAME,  0-VALID;
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});
    auto bias     = NDArrayFactory::create<TypeParam>('c', {oC},{1.f, 2.f, 3.f});
    auto expected = NDArrayFactory::create<TypeParam>('c', {2, 3, 2, 2, 2},{49.f,  49.f, 49.f,  49.f,  49.f,  49.f, 49.f,  49.f,  50.f,  50.f, 50.f,  50.f,  50.f,  50.f, 50.f,  50.f, 
                                                  51.f,  51.f, 51.f,  51.f,  51.f,  51.f, 51.f,  51.f,  49.f,  49.f, 49.f,  49.f,  49.f,  49.f, 49.f,  49.f, 
                                                  50.f,  50.f, 50.f,  50.f,  50.f,  50.f, 50.f,  50.f,  51.f,  51.f, 51.f,  51.f,  51.f,  51.f, 51.f,  51.f});
    input = 2.;
    weights = 0.5;

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    // output->printIndexedBuffer();

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expected.isSameShape(output));
    ASSERT_TRUE(expected.equalsTo(output));

    delete results;
}

////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test7) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int paddingMode = 0;             // 1-SAME,  0-VALID;
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {oC, iC, kD, kH, kW});
    auto bias     = NDArrayFactory::create<TypeParam>('c', {oC},{1.f, 2.f, 3.f});
    auto expected = NDArrayFactory::create<TypeParam>('c', {2, 3, 2, 2, 2},{236.2f, 236.2f, 236.2f, 236.2f, 236.2f, 236.2f, 236.2f, 236.2f, 698.f, 698.f, 698.f, 698.f, 
                                                  698.f, 698.f, 698.f, 698.f, 1159.8f, 1159.8f, 1159.8f, 1159.8f, 1159.8f, 1159.8f, 1159.8f, 1159.8f, 
                                                  236.2f, 236.2f, 236.2f, 236.2f, 236.2f, 236.2f, 236.2f, 236.2f, 698.f, 698.f, 698.f, 698.f, 
                                                  698.f, 698.f, 698.f, 698.f, 1159.8f, 1159.8f, 1159.8f, 1159.8f, 1159.8f, 1159.8f, 1159.8f, 1159.8f});
    input = 2.;
    weights.linspace(0.1, 0.1);
    weights.permutei({2, 3, 4, 1, 0});

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    // output->printIndexedBuffer();

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expected.isSameShape(output));
    ASSERT_TRUE(expected.equalsTo(output));

    delete results;
}

////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test8) {

    int bS=2, iD=3,iH=4,iW=3,  iC=4,oC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int paddingMode = 0;             // 1-SAME,  0-VALID;
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {oC, iC, kD, kH, kW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {2, 3, 2, 2, 2},{235.2f, 235.2f, 235.2f, 235.2f, 235.2f, 235.2f, 235.2f, 235.2f, 696.f, 696.f, 696.f, 696.f, 696.f, 696.f, 696.f, 696.f, 
                                                  1156.8f, 1156.8f, 1156.8f, 1156.8f, 1156.8f, 1156.8f, 1156.8f, 1156.8f, 235.2f, 235.2f, 235.2f, 235.2f, 235.2f, 235.2f, 235.2f, 235.2f, 
                                                  696.f, 696.f, 696.f, 696.f, 696.f, 696.f, 696.f, 696.f, 1156.8f, 1156.8f, 1156.8f, 1156.8f, 1156.8f, 1156.8f, 1156.8f, 1156.8f});
    input = 2.;
    weights.linspace(0.1, 0.1);
    weights.permutei({2, 3, 4, 1, 0});

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expected.isSameShape(output));
    ASSERT_TRUE(expected.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test9) {
    auto x = NDArrayFactory::create<TypeParam>('c', {4, 2, 28, 28, 3});
    auto y = NDArrayFactory::create<TypeParam>('c', {2, 5, 5, 3, 4});
    auto e = NDArrayFactory::create<TypeParam>('c', {4, 1, 7, 10, 4});

    nd4j::ops::conv3dnew op;
    auto result = op.execute({&x, &y}, {}, {2,5,5, 5,4,3, 0,0,0, 1,1,1, 1,1});
    ASSERT_EQ(Status::OK(), result->status());

    auto z = result->at(0);

    ASSERT_TRUE(e.isSameShape(z));

    delete result;
}

TYPED_TEST(TypedConvolutionTests1, conv3d_test10) {
    auto x = NDArrayFactory::create<TypeParam>('c', {4, 2, 28, 28, 3});
    auto w = NDArrayFactory::create<TypeParam>('c', {2, 5, 5, 3, 4});
    auto exp = NDArrayFactory::create<TypeParam>('c', {4, 1, 7, 10, 4});

    nd4j::ops::conv3dnew op;
    auto result = op.execute({&x, &w}, {}, {2,5,5, 5,4,3, 0,0,0, 1,1,1, 1,1});
    ASSERT_EQ(Status::OK(), result->status());

    ShapeList shapeList({x.shapeInfo(), w.shapeInfo()});
    ContextPrototype proto;
    Context ctx(1);
    ctx.getIArguments()->push_back(2);
    ctx.getIArguments()->push_back(5);
    ctx.getIArguments()->push_back(5);

    ctx.getIArguments()->push_back(5);
    ctx.getIArguments()->push_back(4);
    ctx.getIArguments()->push_back(3);

    ctx.getIArguments()->push_back(0);
    ctx.getIArguments()->push_back(0);
    ctx.getIArguments()->push_back(0);

    ctx.getIArguments()->push_back(1);
    ctx.getIArguments()->push_back(1);
    ctx.getIArguments()->push_back(1);

    ctx.getIArguments()->push_back(0);
    ctx.getIArguments()->push_back(1);  // previous variant was "ctx.getIArguments()->push_back(0)" and this caused fail

    auto shapes = op.calculateOutputShape(&shapeList, ctx);
    ASSERT_EQ(1, shapes->size());

    auto s = shapes->at(0);

    auto z = result->at(0);
    // z->printShapeInfo("z shape");

    ASSERT_TRUE(exp.isSameShape(z));

    delete result;

    shapes->destroy();
    delete shapes;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, pointwise_conv2d_test1) {

    int bS=2, iH=4,iW=3,  iC=4,oC=3;

    int dataFormat = 1;           // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {1,   1, iC, oC});
    auto bias     = NDArrayFactory::create<TypeParam>('c', {oC});


    auto expOutput = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, oC},{ 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 
                                                      7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 
                                                      6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 
                                                      5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f, 5.4f, 6.2f, 7.0f});
    input = 2.;
    weights.linspace(0.1, 0.1);
    bias = 1.;

    nd4j::ops::pointwise_conv2d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test11) {

    int bS=1, iD=2,iH=2,iW=2,  iC=1,oC=1,  kD=2,kH=2,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int paddingMode = 1;             // 1-SAME,  0-VALID;
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});

    input = 2.;
    weights = 1.;

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, conv3d_test12) {

    int bS=5, iD=4,iH=14,iW=14,  iC=1,oC=1,  kD=2,kH=2,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int       oD=3,oH=13,oW=13;
    int paddingMode = 0;             // 1-SAME,  0-VALID;
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kD, kH, kW, iC, oC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, oC, oD, oH, oW});

    input = 2.;
    weights = 1.;

    nd4j::ops::conv3dnew op;
    auto results = op.execute({&input, &weights}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(output->isSameShape(&expected));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, vol2col_test1) {

    int bS=2, iD=2,iH=3,iW=2,  iC=3,oC=2,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int       oD=2,oH=3,oW=2;

    NDArray volume('c', {bS, iC, iD, iH, iW}, nd4j::DataType::FLOAT32);
    NDArray columns('c', {bS, iC, kD, kH, kW, oD, oH, oW}, nd4j::DataType::FLOAT32);

    columns = -1.;
    volume.linspace(1);

    NDArray columnsExpected('c', {bS, iC, kD, kH, kW, oD, oH, oW}, {1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11., 12., 2., 0., 4., 0., 6.,0., 8., 0., 10., 0., 12., 0., 3., 4., 5., 6., 0., 0., 9., 10., 11., 12., 0., 0., 4., 0., 6., 0., 0., 0., 10., 0., 12., 0., 0., 0., 5., 6.,
0., 0., 0., 0., 11., 12., 0., 0., 0., 0., 6., 0., 0., 0., 0., 0., 12., 0., 0., 0., 0., 0., 7., 8., 9., 10., 11., 12., 0., 0., 0., 0., 0., 0., 8., 0., 10., 0., 12., 0., 0., 0., 0., 0., 0., 0., 9., 10., 11., 12., 0., 0., 0., 0., 0., 0., 0., 0., 10., 0., 12., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 11., 12., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 12., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 13., 14., 15., 16., 17.,18., 19., 20., 21., 22., 23., 24., 14., 0., 16., 0., 18., 0., 20., 0., 22., 0., 24., 0., 15., 16., 17., 18., 0., 0., 21., 22., 23., 24., 0.,
0., 16., 0., 18., 0., 0., 0., 22., 0., 24., 0., 0., 0., 17., 18., 0., 0., 0., 0., 23., 24., 0., 0., 0., 0., 18., 0., 0., 0., 0., 0., 24., 0., 0., 0., 0., 0., 19., 20., 21., 22., 23., 24., 0., 0., 0., 0., 0., 0., 20., 0., 22., 0., 24., 0., 0., 0., 0., 0., 0., 0., 21., 22., 23.,
24., 0., 0., 0., 0., 0., 0., 0., 0., 22., 0., 24., 0., 0., 0., 0., 0., 0., 0., 0., 0., 23., 24., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 24.,0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 25., 26., 27., 28., 29., 30., 31., 32., 33., 34., 35., 36., 26., 0., 28., 0., 30., 0., 32., 0.,
34., 0., 36., 0., 27., 28., 29., 30., 0., 0., 33., 34., 35., 36., 0., 0., 28., 0., 30., 0., 0., 0., 34., 0., 36., 0., 0., 0., 29., 30., 0., 0., 0., 0., 35., 36., 0., 0., 0., 0., 30., 0., 0., 0., 0., 0., 36., 0., 0., 0., 0., 0., 31., 32., 33., 34., 35., 36., 0., 0., 0., 0., 0.,
0., 32., 0., 34., 0., 36., 0., 0., 0., 0., 0., 0., 0., 33., 34., 35., 36., 0., 0., 0., 0., 0., 0., 0., 0., 34., 0., 36., 0., 0., 0., 0., 0., 0., 0., 0., 0., 35., 36., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 36., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 37., 38., 39., 40.,
41., 42., 43., 44., 45., 46., 47., 48., 38., 0., 40., 0., 42., 0., 44., 0., 46., 0., 48., 0., 39., 40., 41., 42., 0., 0., 45., 46., 47., 48., 0., 0., 40., 0., 42., 0., 0., 0., 46., 0., 48., 0., 0., 0., 41., 42., 0., 0., 0., 0., 47., 48., 0., 0., 0., 0., 42., 0., 0., 0., 0.,
0., 48., 0., 0., 0., 0., 0., 43., 44., 45., 46., 47., 48., 0., 0., 0., 0., 0., 0., 44., 0., 46., 0., 48., 0., 0., 0., 0., 0., 0., 0., 45., 46., 47., 48., 0., 0., 0., 0., 0., 0., 0., 0., 46., 0., 48., 0., 0., 0., 0., 0., 0., 0., 0., 0., 47., 48., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 48., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 49., 50., 51., 52., 53., 54., 55., 56., 57., 58., 59., 60., 50., 0., 52., 0., 54.,0., 56., 0., 58., 0., 60., 0., 51., 52., 53., 54., 0., 0., 57., 58., 59., 60., 0., 0., 52., 0., 54., 0., 0., 0., 58., 0., 60., 0., 0., 0.,
53., 54., 0., 0., 0., 0., 59., 60., 0., 0., 0., 0., 54., 0., 0., 0., 0., 0., 60., 0., 0., 0., 0., 0., 55., 56., 57., 58., 59., 60., 0., 0.,0., 0., 0., 0., 56., 0., 58., 0., 60., 0., 0., 0., 0., 0., 0., 0., 57., 58., 59., 60., 0., 0., 0., 0., 0., 0., 0., 0., 58., 0., 60., 0.,
0., 0., 0., 0., 0., 0., 0., 0., 59., 60., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 60., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 61., 62., 63., 64., 65., 66., 67., 68., 69., 70., 71., 72., 62., 0., 64., 0., 66., 0., 68., 0., 70., 0., 72., 0., 63., 64., 65., 66., 0., 0., 69.,
70., 71., 72., 0., 0., 64., 0., 66., 0., 0., 0., 70., 0., 72., 0., 0., 0., 65., 66., 0., 0., 0., 0., 71., 72., 0., 0., 0., 0., 66., 0., 0., 0., 0., 0., 72., 0., 0., 0., 0., 0., 67., 68., 69., 70., 71., 72., 0., 0., 0., 0., 0., 0., 68., 0., 70., 0., 72., 0., 0., 0., 0., 0., 0.,
0., 69., 70., 71., 72., 0., 0., 0., 0., 0., 0., 0., 0., 70., 0., 72., 0., 0., 0., 0., 0., 0., 0., 0., 0., 71., 72., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 72., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.}, nd4j::DataType::FLOAT32);

    graph::Context context(1);
    nd4j::ops::ConvolutionUtils::vol2col(context, volume, columns, sD, sH, sW, pD, pH, pW, dD, dH, dW);
    // columns.printBuffer();

    ASSERT_TRUE(columns.equalsTo(columnsExpected));
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, vol2col_test2) {

    int bS=2, iD=2,iH=3,iW=2,  iC=3,oC=2,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=1,dH=1,dW=1;
    int       oD=2,oH=3,oW=2;

    auto volume  = NDArrayFactory::create<float>('c', {iD, bS, iH, iC, iW});
    volume.permutei({1, 3, 0, 2, 4});
    volume.linspace(1);

    auto columns = NDArrayFactory::create<float>('c', {kD, iC, kH, oW, kW, bS, oD, oH});
    columns.permutei({5, 1, 0, 2, 4, 6, 7, 3});
    columns = -1.;
    auto columnsExpected = NDArrayFactory::create<float>('c', {bS, iC, kD, kH, kW, oD, oH, oW}, {1.f, 2.f, 3.f, 4.f, 5.f, 6.f, 7.f, 8.f, 9.f, 
10.f, 11.f, 12.f, 2.f, 0.f, 4.f, 0.f, 6.f, 0.f, 8.f, 0.f, 10.f, 0.f, 12.f, 0.f, 3.f, 4.f, 5.f, 6.f, 0.f, 0.f, 9.f, 10.f, 11.f, 12.f, 0.f, 0.f, 4.f, 0.f, 6.f, 0.f, 0.f, 0.f, 10.f, 0.f, 12.f, 0.f, 0.f, 0.f, 5.f, 6.f, 0.f, 0.f, 0.f, 0.f, 11.f, 12.f, 0.f, 0.f, 0.f, 0.f, 6.f, 0.f, 0.f, 0.f, 0.f, 0.f, 12.f, 0.f, 0.f, 0.f, 0.f, 0.f, 7.f, 8.f, 
9.f, 10.f, 11.f, 12.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 8.f, 0.f, 10.f, 0.f, 12.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 9.f, 10.f, 11.f, 12.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 10.f, 0.f, 12.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 11.f, 12.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 12.f, 0.f, 0.f, 0.f, 0.f, 0.f, 
0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 13.f, 14.f, 15.f, 16.f, 17.f, 18.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 14.f, 0.f, 16.f, 0.f, 18.f, 0.f, 20.f, 0.f, 22.f, 0.f, 24.f, 0.f, 15.f, 16.f, 17.f, 18.f, 0.f, 0.f, 21.f, 22.f, 23.f, 24.f, 0.f, 0.f, 16.f, 0.f, 18.f, 0.f, 0.f, 0.f, 22.f, 0.f, 24.f, 0.f, 0.f, 0.f, 17.f, 18.f, 0.f, 0.f, 0.f, 0.f, 
23.f, 24.f, 0.f, 0.f, 0.f, 0.f, 18.f, 0.f, 0.f, 0.f, 0.f, 0.f, 24.f, 0.f, 0.f, 0.f, 0.f, 0.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 20.f, 0.f, 22.f, 0.f, 24.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 21.f, 22.f, 23.f, 24.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 22.f, 0.f, 24.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 
0.f, 0.f, 0.f, 23.f, 24.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 24.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 26.f, 0.f, 28.f, 0.f, 30.f, 0.f, 32.f, 0.f, 34.f, 0.f, 36.f, 0.f, 27.f, 28.f, 29.f, 30.f, 0.f, 0.f, 33.f, 34.f, 35.f, 36.f, 
0.f, 0.f, 28.f, 0.f, 30.f, 0.f, 0.f, 0.f, 34.f, 0.f, 36.f, 0.f, 0.f, 0.f, 29.f, 30.f, 0.f, 0.f, 0.f, 0.f, 35.f, 36.f, 0.f, 0.f, 0.f, 0.f, 30.f, 0.f, 0.f, 0.f, 0.f, 0.f, 36.f, 0.f, 0.f, 0.f, 0.f, 0.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 32.f, 0.f, 34.f, 0.f, 36.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 33.f, 
34.f, 35.f, 36.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 34.f, 0.f, 36.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 35.f, 36.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 36.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 37.f, 38.f, 39.f, 40.f, 41.f, 42.f, 43.f, 44.f, 45.f, 46.f, 47.f, 48.f, 38.f, 0.f, 40.f, 
0.f, 42.f, 0.f, 44.f, 0.f, 46.f, 0.f, 48.f, 0.f, 39.f, 40.f, 41.f, 42.f, 0.f, 0.f, 45.f, 46.f, 47.f, 48.f, 0.f, 0.f, 40.f, 0.f, 42.f, 0.f, 0.f, 0.f, 46.f, 0.f, 48.f, 0.f, 0.f, 0.f, 41.f, 42.f, 0.f, 0.f, 0.f, 0.f, 47.f, 48.f, 0.f, 0.f, 0.f, 0.f, 42.f, 0.f, 0.f, 0.f, 0.f, 0.f, 48.f, 0.f, 0.f, 0.f, 0.f, 0.f, 43.f, 44.f, 45.f, 46.f, 47.f, 
48.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 44.f, 0.f, 46.f, 0.f, 48.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 45.f, 46.f, 47.f, 48.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 46.f, 0.f, 48.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 47.f, 48.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 48.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 
0.f, 0.f, 49.f, 50.f, 51.f, 52.f, 53.f, 54.f, 55.f, 56.f, 57.f, 58.f, 59.f, 60.f, 50.f, 0.f, 52.f, 0.f, 54.f, 0.f, 56.f, 0.f, 58.f, 0.f, 60.f, 0.f, 51.f, 52.f, 53.f, 54.f, 0.f, 0.f, 57.f, 58.f, 59.f, 60.f, 0.f, 0.f, 52.f, 0.f, 54.f, 0.f, 0.f, 0.f, 58.f, 0.f, 60.f, 0.f, 0.f, 0.f, 53.f, 54.f, 0.f, 0.f, 0.f, 0.f, 59.f, 60.f, 0.f, 0.f, 
0.f, 0.f, 54.f, 0.f, 0.f, 0.f, 0.f, 0.f, 60.f, 0.f, 0.f, 0.f, 0.f, 0.f, 55.f, 56.f, 57.f, 58.f, 59.f, 60.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 56.f, 0.f, 58.f, 0.f, 60.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 57.f, 58.f, 59.f, 60.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 58.f, 0.f, 60.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 59.f, 60.f, 
0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 60.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 61.f, 62.f, 63.f, 64.f, 65.f, 66.f, 67.f, 68.f, 69.f, 70.f, 71.f, 72.f, 62.f, 0.f, 64.f, 0.f, 66.f, 0.f, 68.f, 0.f, 70.f, 0.f, 72.f, 0.f, 63.f, 64.f, 65.f, 66.f, 0.f, 0.f, 69.f, 70.f, 71.f, 72.f, 0.f, 0.f, 64.f, 0.f, 66.f, 
0.f, 0.f, 0.f, 70.f, 0.f, 72.f, 0.f, 0.f, 0.f, 65.f, 66.f, 0.f, 0.f, 0.f, 0.f, 71.f, 72.f, 0.f, 0.f, 0.f, 0.f, 66.f, 0.f, 0.f, 0.f, 0.f, 0.f, 72.f, 0.f, 0.f, 0.f, 0.f, 0.f, 67.f, 68.f, 69.f, 70.f, 71.f, 72.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 68.f, 0.f, 70.f, 0.f, 72.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 69.f, 70.f, 71.f, 72.f, 0.f, 0.f, 
0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 70.f, 0.f, 72.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 71.f, 72.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 72.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f});

    graph::Context context(1);
    nd4j::ops::ConvolutionUtils::vol2col(context, volume, columns, sD, sH, sW, pD, pH, pW, dD, dH, dW);
        // columns.printBuffer();

    ASSERT_TRUE(columns.equalsTo(columnsExpected));
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, col2im_test1) {

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

    auto image  = NDArrayFactory::create<float>('c', {bS, iC, iH, iW});
    image = -2.;

    auto columns = NDArrayFactory::create<float>('c', {bS, iC, kH, kW, oH, oW});
    columns.linspace(1);

    auto imageExpected = NDArrayFactory::create<float>('c', {bS, iC, iH, iW}, {1.f,  7.f,  12.f,  34.f,  17.f,  39.f,  44.f,  98.f,  33.f,  71.f,  76.f,  162.f,  49.f,  103.f,  108.f,  226.f});

    LaunchContext ctx;
    nd4j::ops::helpers::col2im(ctx, columns, image, sH, sW, pH, pW, iH, iW, dH, dW);

    ASSERT_TRUE(image.equalsTo(imageExpected));
}


//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, upsampling2d_test1) {

    const int bS=3,  iH=2,iW=2,  iC=3;
    const int factorH=2, factorW=3;
    const int isNCHW = 0;                    // data format, default is NCHW

    auto input  = NDArrayFactory::create<float>('c', {bS, iH, iW, iC});
    input.linspace(1);

    auto expOutput = NDArrayFactory::create<float>('c', {bS, iH*factorH, iW*factorW, iC}, {1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 
                                         7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 
                                        13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 
                                        19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 
                                        25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 
                                        31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f, 31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f});

    nd4j::ops::upsampling2d op;
    auto results = op.execute({&input}, {}, {factorH, factorW, isNCHW});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, upsampling2d_test2) {

    const int bS=3,  iH=2,iW=2,  iC=3;
    const int factorH=2, factorW=3;
    const int isNCHW = 1;                    // data format, default is NCHW

    auto input  = NDArrayFactory::create<float>('c', {bS, iC, iH, iW});
    input.linspace(1);

    auto expOutput = NDArrayFactory::create<float>('c', {bS, iC, iH*factorH, iW*factorW}, {1.f, 1.f, 1.f, 2.f, 2.f, 2.f, 1.f, 1.f, 1.f, 2.f, 2.f, 2.f, 3.f, 3.f, 3.f, 4.f, 4.f, 4.f, 3.f, 3.f, 3.f, 4.f, 4.f, 4.f, 
                                 5.f, 5.f, 5.f, 6.f, 6.f, 6.f, 5.f, 5.f, 5.f, 6.f, 6.f, 6.f, 7.f, 7.f, 7.f, 8.f, 8.f, 8.f, 7.f, 7.f, 7.f, 8.f, 8.f, 8.f, 9.f, 9.f, 9.f, 10.f, 10.f, 10.f, 9.f, 9.f, 9.f, 10.f, 10.f, 10.f, 11.f, 11.f, 11.f, 12.f, 12.f, 12.f, 11.f, 11.f, 11.f, 12.f, 12.f, 12.f, 
                                13.f, 13.f, 13.f, 14.f, 14.f, 14.f, 13.f, 13.f, 13.f, 14.f, 14.f, 14.f, 15.f, 15.f, 15.f, 16.f, 16.f, 16.f, 15.f, 15.f, 15.f, 16.f, 16.f, 16.f, 17.f, 17.f, 17.f, 18.f, 18.f, 18.f, 17.f, 17.f, 17.f, 18.f, 18.f, 18.f, 19.f, 19.f, 19.f, 20.f, 20.f, 20.f, 19.f, 19.f, 19.f, 20.f, 20.f, 20.f, 
                                21.f, 21.f, 21.f, 22.f, 22.f, 22.f, 21.f, 21.f, 21.f, 22.f, 22.f, 22.f, 23.f, 23.f, 23.f, 24.f, 24.f, 24.f, 23.f, 23.f, 23.f, 24.f, 24.f, 24.f, 25.f, 25.f, 25.f, 26.f, 26.f, 26.f, 25.f, 25.f, 25.f, 26.f, 26.f, 26.f, 27.f, 27.f, 27.f, 28.f, 28.f, 28.f, 27.f, 27.f, 27.f, 28.f, 28.f, 28.f, 
                                29.f, 29.f, 29.f, 30.f, 30.f, 30.f, 29.f, 29.f, 29.f, 30.f, 30.f, 30.f, 31.f, 31.f, 31.f, 32.f, 32.f, 32.f, 31.f, 31.f, 31.f, 32.f, 32.f, 32.f, 
                                33.f, 33.f, 33.f, 34.f, 34.f, 34.f, 33.f, 33.f, 33.f, 34.f, 34.f, 34.f, 35.f, 35.f, 35.f, 36.f, 36.f, 36.f, 35.f, 35.f, 35.f, 36.f, 36.f, 36.f});

    nd4j::ops::upsampling2d op;
    auto results = op.execute({&input}, {}, {factorH, factorW, isNCHW});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, upsampling3d_test1) {

    const int bS=3,  iD=2,iH=2,iW=2,  iC=3;
    const int factorD=2,factorH=3,factorW=2;
    const int isNCDHW = 0;                    // data format, default is NCHW

    auto input  = NDArrayFactory::create<float>('c', {bS, iD, iH, iW, iC});
    input.linspace(1);

    auto expOutput = NDArrayFactory::create<float>('c', {bS, iD*factorD, iH*factorH, iW*factorW, iC}, {1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 
             7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 1.f, 2.f, 3.f, 1.f, 2.f, 3.f, 4.f, 5.f, 6.f, 4.f, 5.f, 6.f, 7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 
             7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 7.f, 8.f, 9.f, 7.f, 8.f, 9.f, 10.f, 11.f, 12.f, 10.f, 11.f, 12.f, 13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 
            19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 
            13.f, 14.f, 15.f, 13.f, 14.f, 15.f, 16.f, 17.f, 18.f, 16.f, 17.f, 18.f, 19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 19.f, 20.f, 21.f, 19.f, 20.f, 21.f, 22.f, 23.f, 24.f, 22.f, 23.f, 24.f, 25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 
            25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f, 31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f, 31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f, 
            25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 25.f, 26.f, 27.f, 25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 28.f, 29.f, 30.f, 31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f, 31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f, 
            31.f, 32.f, 33.f, 31.f, 32.f, 33.f, 34.f, 35.f, 36.f, 34.f, 35.f, 36.f, 37.f, 38.f, 39.f, 37.f, 38.f, 39.f, 40.f, 41.f, 42.f, 40.f, 41.f, 42.f, 37.f, 38.f, 39.f, 37.f, 38.f, 39.f, 40.f, 41.f, 42.f, 40.f, 41.f, 42.f, 37.f, 38.f, 39.f, 37.f, 38.f, 39.f, 40.f, 41.f, 42.f, 40.f, 41.f, 42.f, 43.f, 44.f, 45.f, 43.f, 44.f, 45.f, 46.f, 47.f, 48.f, 46.f, 47.f, 48.f, 
            43.f, 44.f, 45.f, 43.f, 44.f, 45.f, 46.f, 47.f, 48.f, 46.f, 47.f, 48.f, 43.f, 44.f, 45.f, 43.f, 44.f, 45.f, 46.f, 47.f, 48.f, 46.f, 47.f, 48.f, 37.f, 38.f, 39.f, 37.f, 38.f, 39.f, 40.f, 41.f, 42.f, 40.f, 41.f, 42.f, 37.f, 38.f, 39.f, 37.f, 38.f, 39.f, 40.f, 41.f, 42.f, 40.f, 41.f, 42.f, 37.f, 38.f, 39.f, 37.f, 38.f, 39.f, 40.f, 41.f, 42.f, 40.f, 41.f, 42.f, 
            43.f, 44.f, 45.f, 43.f, 44.f, 45.f, 46.f, 47.f, 48.f, 46.f, 47.f, 48.f, 43.f, 44.f, 45.f, 43.f, 44.f, 45.f, 46.f, 47.f, 48.f, 46.f, 47.f, 48.f, 43.f, 44.f, 45.f, 43.f, 44.f, 45.f, 46.f, 47.f, 48.f, 46.f, 47.f, 48.f, 49.f, 50.f, 51.f, 49.f, 50.f, 51.f, 52.f, 53.f, 54.f, 52.f, 53.f, 54.f, 49.f, 50.f, 51.f, 49.f, 50.f, 51.f, 52.f, 53.f, 54.f, 52.f, 53.f, 54.f, 
            49.f, 50.f, 51.f, 49.f, 50.f, 51.f, 52.f, 53.f, 54.f, 52.f, 53.f, 54.f, 55.f, 56.f, 57.f, 55.f, 56.f, 57.f, 58.f, 59.f, 60.f, 58.f, 59.f, 60.f, 55.f, 56.f, 57.f, 55.f, 56.f, 57.f, 58.f, 59.f, 60.f, 58.f, 59.f, 60.f, 55.f, 56.f, 57.f, 55.f, 56.f, 57.f, 58.f, 59.f, 60.f, 58.f, 59.f, 60.f, 49.f, 50.f, 51.f, 49.f, 50.f, 51.f, 52.f, 53.f, 54.f, 52.f, 53.f, 54.f, 
            49.f, 50.f, 51.f, 49.f, 50.f, 51.f, 52.f, 53.f, 54.f, 52.f, 53.f, 54.f, 49.f, 50.f, 51.f, 49.f, 50.f, 51.f, 52.f, 53.f, 54.f, 52.f, 53.f, 54.f, 55.f, 56.f, 57.f, 55.f, 56.f, 57.f, 58.f, 59.f, 60.f, 58.f, 59.f, 60.f, 55.f, 56.f, 57.f, 55.f, 56.f, 57.f, 58.f, 59.f, 60.f, 58.f, 59.f, 60.f, 55.f, 56.f, 57.f, 55.f, 56.f, 57.f, 58.f, 59.f, 60.f, 58.f, 59.f, 60.f, 
            61.f, 62.f, 63.f, 61.f, 62.f, 63.f, 64.f, 65.f, 66.f, 64.f, 65.f, 66.f, 61.f, 62.f, 63.f, 61.f, 62.f, 63.f, 64.f, 65.f, 66.f, 64.f, 65.f, 66.f, 61.f, 62.f, 63.f, 61.f, 62.f, 63.f, 64.f, 65.f, 66.f, 64.f, 65.f, 66.f, 67.f, 68.f, 69.f, 67.f, 68.f, 69.f, 70.f, 71.f, 72.f, 70.f, 71.f, 72.f, 67.f, 68.f, 69.f, 67.f, 68.f, 69.f, 70.f, 71.f, 72.f, 70.f, 71.f, 72.f, 
            67.f, 68.f, 69.f, 67.f, 68.f, 69.f, 70.f, 71.f, 72.f, 70.f, 71.f, 72.f, 61.f, 62.f, 63.f, 61.f, 62.f, 63.f, 64.f, 65.f, 66.f, 64.f, 65.f, 66.f, 61.f, 62.f, 63.f, 61.f, 62.f, 63.f, 64.f, 65.f, 66.f, 64.f, 65.f, 66.f, 61.f, 62.f, 63.f, 61.f, 62.f, 63.f, 64.f, 65.f, 66.f, 64.f, 65.f, 66.f, 67.f, 68.f, 69.f, 67.f, 68.f, 69.f, 70.f, 71.f, 72.f, 70.f, 71.f, 72.f, 
            67.f, 68.f, 69.f, 67.f, 68.f, 69.f, 70.f, 71.f, 72.f, 70.f, 71.f, 72.f, 67.f, 68.f, 69.f, 67.f, 68.f, 69.f, 70.f, 71.f, 72.f, 70.f, 71.f, 72.f});

    nd4j::ops::upsampling3d op;
    auto results = op.execute({&input}, {}, {factorD, factorH, factorW, isNCDHW});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, upsampling3d_test2) {

    const int bS=3,  iD=2,iH=2,iW=2,  iC=3;
    const int factorD=2,factorH=3,factorW=2;
    const int isNCDHW = 1;                    // data format, default is NCHW

    auto input = NDArrayFactory::create<float>('c', {bS, iC, iD, iH, iW});
    input.linspace(1);

    auto expOutput = NDArrayFactory::create<float>('c', {bS, iC, iD*factorD, iH*factorH, iW*factorW}, { 1.f, 1.f, 2.f, 2.f, 1.f, 1.f, 2.f, 2.f, 1.f, 1.f, 2.f, 2.f, 3.f, 3.f, 4.f, 4.f, 3.f, 3.f, 4.f, 4.f, 3.f, 3.f, 4.f, 4.f, 1.f, 1.f, 2.f, 2.f, 1.f, 1.f, 2.f, 2.f, 1.f, 1.f, 2.f, 2.f, 3.f, 3.f, 4.f, 4.f, 3.f, 3.f, 4.f, 4.f, 3.f, 3.f, 4.f, 4.f, 5.f, 5.f, 6.f, 6.f, 5.f, 5.f, 6.f, 6.f, 5.f, 5.f, 6.f, 6.f, 7.f, 7.f, 8.f, 8.f, 7.f, 7.f, 8.f, 8.f, 7.f, 7.f, 8.f, 8.f, 
             5.f, 5.f, 6.f, 6.f, 5.f, 5.f, 6.f, 6.f, 5.f, 5.f, 6.f, 6.f, 7.f, 7.f, 8.f, 8.f, 7.f, 7.f, 8.f, 8.f, 7.f, 7.f, 8.f, 8.f, 9.f, 9.f, 10.f, 10.f, 9.f, 9.f, 10.f, 10.f, 9.f, 9.f, 10.f, 10.f, 11.f, 11.f, 12.f, 12.f, 11.f, 11.f, 12.f, 12.f, 11.f, 11.f, 12.f, 12.f, 9.f, 9.f, 10.f, 10.f, 9.f, 9.f, 10.f, 10.f, 9.f, 9.f, 10.f, 10.f, 11.f, 11.f, 12.f, 12.f, 11.f, 11.f, 12.f, 12.f, 11.f, 11.f, 12.f, 12.f, 
            13.f, 13.f, 14.f, 14.f, 13.f, 13.f, 14.f, 14.f, 13.f, 13.f, 14.f, 14.f, 15.f, 15.f, 16.f, 16.f, 15.f, 15.f, 16.f, 16.f, 15.f, 15.f, 16.f, 16.f, 13.f, 13.f, 14.f, 14.f, 13.f, 13.f, 14.f, 14.f, 13.f, 13.f, 14.f, 14.f, 15.f, 15.f, 16.f, 16.f, 15.f, 15.f, 16.f, 16.f, 15.f, 15.f, 16.f, 16.f, 17.f, 17.f, 18.f, 18.f, 17.f, 17.f, 18.f, 18.f, 17.f, 17.f, 18.f, 18.f, 19.f, 19.f, 20.f, 20.f, 19.f, 19.f, 20.f, 20.f, 19.f, 19.f, 20.f, 20.f, 
            17.f, 17.f, 18.f, 18.f, 17.f, 17.f, 18.f, 18.f, 17.f, 17.f, 18.f, 18.f, 19.f, 19.f, 20.f, 20.f, 19.f, 19.f, 20.f, 20.f, 19.f, 19.f, 20.f, 20.f, 21.f, 21.f, 22.f, 22.f, 21.f, 21.f, 22.f, 22.f, 21.f, 21.f, 22.f, 22.f, 23.f, 23.f, 24.f, 24.f, 23.f, 23.f, 24.f, 24.f, 23.f, 23.f, 24.f, 24.f, 21.f, 21.f, 22.f, 22.f, 21.f, 21.f, 22.f, 22.f, 21.f, 21.f, 22.f, 22.f, 23.f, 23.f, 24.f, 24.f, 23.f, 23.f, 24.f, 24.f, 23.f, 23.f, 24.f, 24.f, 
            25.f, 25.f, 26.f, 26.f, 25.f, 25.f, 26.f, 26.f, 25.f, 25.f, 26.f, 26.f, 27.f, 27.f, 28.f, 28.f, 27.f, 27.f, 28.f, 28.f, 27.f, 27.f, 28.f, 28.f, 25.f, 25.f, 26.f, 26.f, 25.f, 25.f, 26.f, 26.f, 25.f, 25.f, 26.f, 26.f, 27.f, 27.f, 28.f, 28.f, 27.f, 27.f, 28.f, 28.f, 27.f, 27.f, 28.f, 28.f, 29.f, 29.f, 30.f, 30.f, 29.f, 29.f, 30.f, 30.f, 29.f, 29.f, 30.f, 30.f, 31.f, 31.f, 32.f, 32.f, 31.f, 31.f, 32.f, 32.f, 31.f, 31.f, 32.f, 32.f, 
            29.f, 29.f, 30.f, 30.f, 29.f, 29.f, 30.f, 30.f, 29.f, 29.f, 30.f, 30.f, 31.f, 31.f, 32.f, 32.f, 31.f, 31.f, 32.f, 32.f, 31.f, 31.f, 32.f, 32.f, 33.f, 33.f, 34.f, 34.f, 33.f, 33.f, 34.f, 34.f, 33.f, 33.f, 34.f, 34.f, 35.f, 35.f, 36.f, 36.f, 35.f, 35.f, 36.f, 36.f, 35.f, 35.f, 36.f, 36.f, 33.f, 33.f, 34.f, 34.f, 33.f, 33.f, 34.f, 34.f, 33.f, 33.f, 34.f, 34.f, 35.f, 35.f, 36.f, 36.f, 35.f, 35.f, 36.f, 36.f, 35.f, 35.f, 36.f, 36.f, 
            37.f, 37.f, 38.f, 38.f, 37.f, 37.f, 38.f, 38.f, 37.f, 37.f, 38.f, 38.f, 39.f, 39.f, 40.f, 40.f, 39.f, 39.f, 40.f, 40.f, 39.f, 39.f, 40.f, 40.f, 37.f, 37.f, 38.f, 38.f, 37.f, 37.f, 38.f, 38.f, 37.f, 37.f, 38.f, 38.f, 39.f, 39.f, 40.f, 40.f, 39.f, 39.f, 40.f, 40.f, 39.f, 39.f, 40.f, 40.f, 41.f, 41.f, 42.f, 42.f, 41.f, 41.f, 42.f, 42.f, 41.f, 41.f, 42.f, 42.f, 43.f, 43.f, 44.f, 44.f, 43.f, 43.f, 44.f, 44.f, 43.f, 43.f, 44.f, 44.f, 
            41.f, 41.f, 42.f, 42.f, 41.f, 41.f, 42.f, 42.f, 41.f, 41.f, 42.f, 42.f, 43.f, 43.f, 44.f, 44.f, 43.f, 43.f, 44.f, 44.f, 43.f, 43.f, 44.f, 44.f, 45.f, 45.f, 46.f, 46.f, 45.f, 45.f, 46.f, 46.f, 45.f, 45.f, 46.f, 46.f, 47.f, 47.f, 48.f, 48.f, 47.f, 47.f, 48.f, 48.f, 47.f, 47.f, 48.f, 48.f, 45.f, 45.f, 46.f, 46.f, 45.f, 45.f, 46.f, 46.f, 45.f, 45.f, 46.f, 46.f, 47.f, 47.f, 48.f, 48.f, 47.f, 47.f, 48.f, 48.f, 47.f, 47.f, 48.f, 48.f, 
            49.f, 49.f, 50.f, 50.f, 49.f, 49.f, 50.f, 50.f, 49.f, 49.f, 50.f, 50.f, 51.f, 51.f, 52.f, 52.f, 51.f, 51.f, 52.f, 52.f, 51.f, 51.f, 52.f, 52.f, 49.f, 49.f, 50.f, 50.f, 49.f, 49.f, 50.f, 50.f, 49.f, 49.f, 50.f, 50.f, 51.f, 51.f, 52.f, 52.f, 51.f, 51.f, 52.f, 52.f, 51.f, 51.f, 52.f, 52.f, 53.f, 53.f, 54.f, 54.f, 53.f, 53.f, 54.f, 54.f, 53.f, 53.f, 54.f, 54.f, 55.f, 55.f, 56.f, 56.f, 55.f, 55.f, 56.f, 56.f, 55.f, 55.f, 56.f, 56.f, 
            53.f, 53.f, 54.f, 54.f, 53.f, 53.f, 54.f, 54.f, 53.f, 53.f, 54.f, 54.f, 55.f, 55.f, 56.f, 56.f, 55.f, 55.f, 56.f, 56.f, 55.f, 55.f, 56.f, 56.f, 57.f, 57.f, 58.f, 58.f, 57.f, 57.f, 58.f, 58.f, 57.f, 57.f, 58.f, 58.f, 59.f, 59.f, 60.f, 60.f, 59.f, 59.f, 60.f, 60.f, 59.f, 59.f, 60.f, 60.f, 57.f, 57.f, 58.f, 58.f, 57.f, 57.f, 58.f, 58.f, 57.f, 57.f, 58.f, 58.f, 59.f, 59.f, 60.f, 60.f, 59.f, 59.f, 60.f, 60.f, 59.f, 59.f, 60.f, 60.f, 
            61.f, 61.f, 62.f, 62.f, 61.f, 61.f, 62.f, 62.f, 61.f, 61.f, 62.f, 62.f, 63.f, 63.f, 64.f, 64.f, 63.f, 63.f, 64.f, 64.f, 63.f, 63.f, 64.f, 64.f, 61.f, 61.f, 62.f, 62.f, 61.f, 61.f, 62.f, 62.f, 61.f, 61.f, 62.f, 62.f, 63.f, 63.f, 64.f, 64.f, 63.f, 63.f, 64.f, 64.f, 63.f, 63.f, 64.f, 64.f, 65.f, 65.f, 66.f, 66.f, 65.f, 65.f, 66.f, 66.f, 65.f, 65.f, 66.f, 66.f, 67.f, 67.f, 68.f, 68.f, 67.f, 67.f, 68.f, 68.f, 67.f, 67.f, 68.f, 68.f, 
            65.f, 65.f, 66.f, 66.f, 65.f, 65.f, 66.f, 66.f, 65.f, 65.f, 66.f, 66.f, 67.f, 67.f, 68.f, 68.f, 67.f, 67.f, 68.f, 68.f, 67.f, 67.f, 68.f, 68.f, 69.f, 69.f, 70.f, 70.f, 69.f, 69.f, 70.f, 70.f, 69.f, 69.f, 70.f, 70.f, 71.f, 71.f, 72.f, 72.f, 71.f, 71.f, 72.f, 72.f, 71.f, 71.f, 72.f, 72.f, 69.f, 69.f, 70.f, 70.f, 69.f, 69.f, 70.f, 70.f, 69.f, 69.f, 70.f, 70.f, 71.f, 71.f, 72.f, 72.f, 71.f, 71.f, 72.f, 72.f, 71.f, 71.f, 72.f, 72.f});

    nd4j::ops::upsampling3d op;
    auto results = op.execute({&input}, {}, {factorD, factorH, factorW, isNCDHW});
    auto* output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output));

    delete results;
}


//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, upsampling3d_bp_test1) {

    const int bS=1,  iD=2,iH=2,iW=2,  iC=1;
    const int factorD=2, factorH=2, factorW=2;
    const int isNCDHW = 1;                    // data format, default is NCHW

    auto input  = NDArrayFactory::create<float>('c', {bS, iC, iD, iH, iW});
    auto gradO  = NDArrayFactory::create<float>('c', {bS, iC, iD*factorD, iH*factorH, iW*factorW});
    gradO = 1.;

    auto expGradI = NDArrayFactory::create<float>('c', {bS, iC, iD, iH, iW});
    expGradI = 8.;

    nd4j::ops::upsampling3d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {isNCDHW});
    auto* gradI = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    delete results;
}

TYPED_TEST(TypedConvolutionTests1, conv2D_input_BP_test1) {

    auto inputShape = NDArrayFactory::create<TypeParam>('c', {4}, {2, 1, 4, 4});
    auto weights = NDArrayFactory::create<TypeParam>('c', {2, 1, 3, 3});
    auto epsilonNext = NDArrayFactory::create<TypeParam>('c', {2, 2, 4, 4});
    auto shapeArr = NDArrayFactory::create<TypeParam>('c', {2, 1, 4, 4});


    TypeParam _expEpsB[] = {952.0, 1540.0, 1636.0, 1180.0, 1791.0, 2886.0, 3057.0, 2193.0, 2223.0, 3570.0, 3741.0, 2673.0, 1900.0, 3028.0, 3160.0, 2240.0, 2872.0, 4612.0, 4708.0, 3356.0, 5247.0, 8358.0, 8529.0, 6033.0, 5679.0, 9042.0, 9213.0, 6513.0, 4588.0, 7252.0, 7384.0, 5184.0};
    NDArray expEps(_expEpsB, shapeArr.getShapeInfo());

    weights.linspace(1);
    epsilonNext.linspace(1);
    weights.permutei({2,3,1,0});

    nd4j::ops::conv2d_input_bp op;

    auto results = op.execute({&inputShape, &weights, &epsilonNext}, {},  {3, 3, 1, 1, 0, 0, 1, 1, 1});

    ASSERT_TRUE(results->size() == 1);

    auto epsilon = results->at(0);

    ASSERT_TRUE(shapeArr.isSameShape(epsilon));
    ASSERT_TRUE(expEps.equalsTo(epsilon));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, upsampling3d_bp_test3) {

    const int bS=1,  iD=3,iH=3,iW=3,  iC=2;
    const int factorD=2, factorH=2, factorW=2;
    const int isNCDHW = 1;                    // data format, default is NCHW

    NDArray input('c', {bS, iC, iD, iH, iW}, nd4j::DataType::FLOAT32);
    NDArray gradO('c', {bS, iC, iD*factorD, iH*factorH, iW*factorW}, {0.6793504, 0.35508695, 0.84278935, 0.20031333, 0.7014987, 0.31069338,
        0.44793984, 0.93800974, 0.32667395, 0.15187258, 0.38331753, 0.78212297, 0.1988072, 0.7985636, 0.1632634, 0.14696825, 0.26089668,
        0.13505761, 0.7562093, 0.27545404, 0.36908787, 0.09282647, 0.83649176, 0.26841334, 0.09506222, 0.31279507, 0.13591796, 0.5175439,
        0.32870287, 0.061735712, 0.39643127, 0.248016, 0.5489592, 0.115046196, 0.8143622, 0.7215636, 0.40449402, 0.29908907, 0.4038839,
        0.9883108, 0.022296403, 0.927782, 0.3184157, 0.0685462, 0.28453344, 0.23272, 0.35214192, 0.058909304, 0.7112212, 0.6744568, 0.19694561,
        0.6994972, 0.0743224, 0.42042503, 0.5842631, 0.14957358, 0.44640633, 0.72307247, 0.06448108, 0.48307765, 0.8759956, 0.5698191, 0.4458631,
        0.5277549, 0.016646361, 0.753678, 0.14063567, 0.7541292, 0.16193217, 0.7750374, 0.3326449, 0.11739397, 0.017710684, 0.60847557, 0.52515227,
        0.9171938, 0.84989065, 0.5894228, 0.85227835, 0.39063585, 0.88968325, 0.6694452, 0.698873, 0.96147966, 0.15740126, 0.15736352, 0.49352047,
        0.5699365, 0.12683152, 0.11572781, 0.7863682, 0.737939, 0.49007934, 0.6084143, 0.9564999, 0.3900982, 0.14730452, 0.8506447, 0.49765033,
        0.07186628, 0.08214969, 0.035314173, 0.7320408, 0.36993408, 0.8406658, 0.27389422, 0.43179566, 0.13323106, 0.19297548, 0.24689731, 0.38641843,
        0.51154125, 0.19903564, 0.1416313, 0.69769853, 0.25363067, 0.78221816, 0.9300991, 0.3355119, 0.5588076, 0.6643576, 0.018850708, 0.63755876,
        0.2904297, 0.43490165, 0.84251267, 0.46609768, 0.38139546, 0.52318525, 0.9901826, 0.9257676, 0.6434591, 0.016828254, 0.9187561, 0.22897908,
        0.0063138064, 0.66597503, 0.19036093, 0.59552056, 0.69888055, 0.22146936, 0.9124342, 0.8708221, 0.7273687, 0.52397245, 0.66288394, 0.2188415,
        0.3354802, 0.03566524, 0.5101009, 0.5017283, 0.75122046, 0.1884508, 0.7407126, 0.6253045, 0.47145858, 0.5369367, 0.19884548, 0.99008304,
        0.08256686, 0.91884845, 0.02360027, 0.98895234, 0.3751719, 0.91783875, 0.4338776, 0.6783008, 0.6667967, 0.46720362, 0.7508773, 0.52304846,
        0.76631916, 0.4187526, 0.7653719, 0.5159193, 0.42730415, 0.49462363, 0.2731735, 0.8862948, 0.043214794, 0.3197591, 0.040378205, 0.5427239,
        0.9228089, 0.045940384, 0.70047987, 0.8419288, 0.53966296, 0.009444186, 0.038044546, 0.03158029, 0.43485752, 0.9204235, 0.5478789, 0.8290083,
        0.11868837, 0.0229866, 0.6639305, 0.8757367, 0.8279557, 0.76270294, 0.43242732, 0.4713431, 0.2569212, 0.30575937, 0.44395888, 0.99384075,
        0.6127142, 0.44844577, 0.6347944, 0.098358564, 0.34233716, 0.9329664, 0.65776783, 0.108565055, 0.2052629, 0.46441218, 0.041791342, 0.89369565,
        0.7000381, 0.2106213, 0.51152664, 0.44200692, 0.8293282, 0.20901772, 0.6387249, 0.8016979, 0.11178707, 0.109545894, 0.19654618, 0.060582615,
        0.08239174, 0.64630795, 0.32862368, 0.60225064, 0.8328141, 0.5484566, 0.8120276, 0.38822946, 0.6742381, 0.34913155, 0.42887798, 0.45344824,
        0.73956585, 0.9714739, 0.42937812, 0.45185348, 0.84535813, 0.046436775, 0.8802151, 0.8676222, 0.42625394, 0.4985318, 0.42399272, 0.122144565,
        0.0060101906, 0.47253844, 0.18123977, 0.86316174, 0.5863874, 0.3852012, 0.9785553, 0.0054711984, 0.88500834, 0.020897374, 0.27467912, 0.3852802,
        0.0766939, 0.94622654, 0.38687763, 0.3308602, 0.7770494, 0.9052543, 0.22258204, 0.42207044, 0.18050623, 0.21057767, 0.012561422, 0.7977821,
        0.61251044, 0.7203693, 0.6028265, 0.6036933, 0.1446382, 0.6712341, 0.76634467, 0.4854034, 0.26634562, 0.76523924, 0.16348523, 0.2663676,
        0.96846986, 0.8273284, 0.10700377, 0.7600526, 0.6771002, 0.47963092, 0.21264452, 0.56934077, 0.5514792, 0.85725874, 0.99090636, 0.54562527,
        0.93597686, 0.21142527, 0.4628326, 0.35011524, 0.31464386, 0.31164807, 0.65928996, 0.94418925, 0.39666295, 0.9496393, 0.103756346, 0.482158,
        0.49171793, 0.4108867, 0.22594318, 0.97093135, 0.5974685, 0.34632966, 0.54835194, 0.10499302, 0.9767778, 0.55008715, 0.54379046, 0.3583731,
        0.33369112, 0.04279039, 0.24939054, 0.23943715, 0.06775989, 0.7750291, 0.24329625, 0.4327169, 0.86916673, 0.80322117, 0.049972698, 0.47177452,
        0.37419558, 0.15303156, 0.121425234, 0.75884604, 0.8191354, 0.48554084, 0.053899214, 0.7858246, 0.39219773, 0.77579063, 0.34507045, 0.46070176,
        0.14496958, 0.47706795, 0.50678796, 0.64902323, 0.3277943, 0.0017530271, 0.6536156, 0.8582253, 0.95703506, 0.9963951, 0.8239163, 0.305142,
        0.012419582, 0.9498972, 0.1595827, 0.47947606, 0.5071124, 0.78227425, 0.2066719, 0.5217094, 0.7841406, 0.5260441, 0.49798164, 0.10975622,
        0.8633349, 0.76298475, 0.14295428, 0.6131504, 0.43794408, 0.50339264, 0.4504877, 0.19235311, 0.6678411, 0.80769485, 0.67495126, 0.96461457,
        0.10535406, 0.66438645, 0.4372345, 0.93851465, 0.8635335, 0.3405871, 0.45652762, 0.3636232, 0.52931345, 0.20154329, 0.07698499, 0.6125804,
        0.3583082, 0.3894796, 0.32601944, 0.5237369, 0.66683626, 0.08541841, 0.4815708, 0.11897489, 0.97555137, 0.3602705, 0.9620871, 0.6361821,
        0.71167386, 0.5134439, 0.57761437, 0.58598644, 0.39387667, 0.6966405, 0.46841687, 0.85788506, 0.9957087, 0.051309288, 0.24846801, 0.55938333,
        0.10230542, 0.9370694, 0.57527155, 0.54656035, 0.28896323, 0.51303476, 0.8865, 0.38641605, 0.9836358}, nd4j::DataType::FLOAT32);

    NDArray expGradI('c', {bS, iC, iD, iH, iW}, {3.510932, 3.4310975, 3.538762, 4.148549, 2.8380678, 2.5431657, 3.3928843, 3.228055, 3.1467278,
        3.2603023, 5.611751, 4.334653, 3.3697734, 4.603307, 4.4357986, 4.32991, 3.0532732, 3.1370173, 4.181534, 2.9965065, 2.8553872, 5.2719016,
        4.5671935, 3.7027276, 3.3517184, 5.2544537, 3.5107024, 4.1496124, 3.9333878, 3.1798909, 3.1446428, 3.0932689, 3.9730802, 3.0466917,
        4.9675374, 4.769673, 3.766952, 3.6375027, 3.6492167, 4.9440994, 3.8379507, 3.467589, 4.719474, 3.1295977, 4.5177174, 4.2760015, 2.8443856,
        4.225355, 4.377341, 4.4398847, 4.710785, 4.4199953, 3.928307, 4.8769503}, nd4j::DataType::FLOAT32);

    nd4j::ops::upsampling3d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {isNCDHW});
    auto* gradI = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(expGradI.isSameShape(gradI));
    ASSERT_TRUE(expGradI.equalsTo(gradI));

    delete results;
}


//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, deconv2d_test1) {

    int bS=2, oH=4,oW=4,  oC=5,iC=10,  kH=2,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       iH=3,iW=3;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<float>('c', {bS, iH, iW, iC});
    auto weights  = NDArrayFactory::create<float>('c', {kH, kW, oC, iC});
    auto exp = NDArrayFactory::create<float>('c', {bS, oH, oW, oC}, {  2.75f, 7.75f, 12.75f, 17.75f, 22.75f, 30.5f, 40.5f, 50.5f, 60.5f, 70.5f, 30.5f, 40.5f, 50.5f, 60.5f, 70.5f, 27.75f, 32.75f, 37.75f, 42.75f, 47.75f, 
                                                  55.5f, 65.5f, 75.5f, 85.5f, 95.5f, 161.f, 181.f, 201.f, 221.f, 241.f, 161.f, 181.f, 201.f, 221.f, 241.f, 105.5f, 115.5f, 125.5f, 135.5f, 145.5f, 
                                                  55.5f, 65.5f, 75.5f, 85.5f, 95.5f, 161.f, 181.f, 201.f, 221.f, 241.f, 161.f, 181.f, 201.f, 221.f, 241.f, 105.5f, 115.5f, 125.5f, 135.5f, 145.5f, 
                                                  52.75f, 57.75f, 62.75f, 67.75f, 72.75f, 130.5f, 140.5f, 150.5f, 160.5f, 170.5f, 130.5f, 140.5f, 150.5f, 160.5f, 170.5f, 77.75f, 82.75f, 87.75f, 92.75f, 97.75f, 
                                                   2.75f, 7.75f, 12.75f, 17.75f, 22.75f, 30.5f, 40.5f, 50.5f, 60.5f, 70.5f, 30.5f, 40.5f, 50.5f, 60.5f, 70.5f, 27.75f, 32.75f, 37.75f, 42.75f, 47.75f, 
                                                  55.5f, 65.5f, 75.5f, 85.5f, 95.5f, 161.f, 181.f, 201.f, 221.f, 241.f, 161.f, 181.f, 201.f, 221.f, 241.f, 105.5f, 115.5f, 125.5f, 135.5f, 145.5f, 
                                                  55.5f, 65.5f, 75.5f, 85.5f, 95.5f, 161.f, 181.f, 201.f, 221.f, 241.f, 161.f, 181.f, 201.f, 221.f, 241.f, 105.5f, 115.5f, 125.5f, 135.5f, 145.5f, 
                                                  52.75f, 57.75f, 62.75f, 67.75f, 72.75f, 130.5f, 140.5f, 150.5f, 160.5f, 170.5f, 130.5f, 140.5f, 150.5f, 160.5f, 170.5f, 77.75f, 82.75f, 87.75f, 92.75f, 97.75f});
    input = 0.5;
    weights.linspace(0.1, 0.1);

    nd4j::ops::deconv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    ASSERT_EQ(Status::OK(), results->status());

    auto output = results->at(0);

    ASSERT_TRUE(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, deconv2d_test2) {

    int bS=2, iH=4,iW=4,  iC=5,oC=10,  kH=2,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=4,oW=4;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<float>('c', {bS, oH, oW, oC});
    auto weights  = NDArrayFactory::create<float>('c', {kH, kW, iC, oC});
    auto exp = NDArrayFactory::create<float>('c', {bS, iH, iW, iC}, {2.75f,    7.75f,   12.75f,   17.75f,   22.75f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f, 
                                                55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 
                                                55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 
                                                55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 
                                                 2.75f,    7.75f,   12.75f,   17.75f,   22.75f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f, 
                                                55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 
                                                55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 
                                                55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f  });
    input = 0.5;
    weights.linspace(0.1, 0.1);

    nd4j::ops::deconv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, deconv2d_test3) {

    int bS=1, oH=5,oW=5,  oC=3,iC=2,  kH=2,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=2,dW=2;
    int       iH=3,iW=3;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<float>('c', {bS, iH, iW, iC});
    auto weights  = NDArrayFactory::create<float>('c', {kH, kW, oC, iC});
    auto bias     = NDArrayFactory::create<float>('c', {oC});

    auto exp = NDArrayFactory::create<float>('c', {bS, oH, oW, oC}, {-2.9f, -6.8f, -10.7f, -2.6f, -6.1f, -9.6f, -16.9f, -23.9f, -30.9f, -13.1f, -16.6f, -20.1f, -11.6f, -14.7f, -17.8f, -2.0f, -4.7f, -7.4f, -1.7f, -4.0f, -6.3f, -11.5f, -16.1f,
                                -20.7f, -8.6f, -10.9f, -13.2f, -7.1f, -9.0f, -10.9f, -27.4f, -32.8f, -38.2f, -24.4f, -29.0f, -33.6f, -65.0f, -74.2f, -83.4f, -38.2f, -42.8f, -47.4f,
                                -32.8f, -36.6f, -40.4f, -18.2f, -20.9f, -23.6f, -15.5f, -17.8f, -20.1f, -39.1f, -43.7f, -48.3f, -22.4f, -24.7f, -27.0f, -18.5f, -20.4f, -22.3f, -10.1f, -11.6f, -13.1f,
                                -7.4f, -8.5f, -9.6f, -19.3f, -21.5f, -23.7f, -10.7f, -11.8f, -12.9f, -6.8f, -7.5f, -8.2f});

    input.linspace(-10, 0.5);
    weights.linspace(0.1, 0.1);
    bias = 0.2;

    nd4j::ops::deconv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    ASSERT_EQ(Status::OK(), results->status());

    auto output = results->at(0);

    ASSERT_TRUE(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, deconv2d_test4) {

    NDArray input('c', {2, 3, 4, 4}, nd4j::DataType::FLOAT32);
    NDArray weights('c', {3, 3, 5, 5}, nd4j::DataType::FLOAT32);
    NDArray exp('c', {2,3,8,8}, {6276.0,12831.0,19668.0,26790.0,27012.0,20703.0,14100.0,7200.0,13719.0,28023.0,42918.0,58410.0,58902.0,45105.0,30693.0,15660.0,22389.0,45696.0,69930.0,95100.0,95910.0,73386.0,49899.0,25440.0,32346.0,65970.0,
                                100884.0,137100.0,138276.0,105726.0,71838.0,36600.0,33726.0,68790.0,105204.0,142980.0,144156.0,110226.0,74898.0,38160.0,27555.0,56154.0,85806.0,116520.0,117474.0,89748.0,60933.0,31020.0,19917.0,40557.0,61926.0,
                                84030.0,84714.0,64671.0,43875.0,22320.0,10752.0,21879.0,33384.0,45270.0,45636.0,34815.0,23604.0,12000.0,7551.0,15456.0,23718.0,32340.0,32562.0,24978.0,17025.0,8700.0,16569.0,33873.0,51918.0,70710.0,71202.0,
                                54555.0,37143.0,18960.0,27114.0,55371.0,84780.0,115350.0,116160.0,88911.0,60474.0,30840.0,39246.0,80070.0,122484.0,166500.0,167676.0,128226.0,87138.0,44400.0,40626.0,82890.0,126804.0,172380.0,173556.0,132726.0,
                                90198.0,45960.0,33180.0,67629.0,103356.0,140370.0,141324.0,107973.0,73308.0,37320.0,23967.0,48807.0,74526.0,101130.0,101814.0,77721.0,52725.0,26820.0,12927.0,26304.0,40134.0,54420.0,54786.0,41790.0,28329.0,14400.0,
                                8826.0,18081.0,27768.0,37890.0,38112.0,29253.0,19950.0,10200.0,19419.0,39723.0,60918.0,83010.0,83502.0,64005.0,43593.0,22260.0,31839.0,65046.0,99630.0,135600.0,136410.0,104436.0,71049.0,36240.0,46146.0,94170.0,
                                144084.0,195900.0,197076.0,150726.0,102438.0,52200.0,47526.0,96990.0,148404.0,201780.0,202956.0,155226.0,105498.0,53760.0,38805.0,79104.0,120906.0,164220.0,165174.0,126198.0,85683.0,43620.0,28017.0,57057.0,87126.0,
                                118230.0,118914.0,90771.0,61575.0,31320.0,15102.0,30729.0,46884.0,63570.0,63936.0,48765.0,33054.0,16800.0,17220.0,34863.0,52932.0,71430.0,72228.0,54831.0,36996.0,18720.0,36327.0,73527.0,111606.0,150570.0,152214.0,
                                115521.0,77925.0,39420.0,57381.0,116112.0,176202.0,237660.0,240198.0,182250.0,122907.0,62160.0,80442.0,162738.0,246900.0,332940.0,336420.0,255198.0,172062.0,87000.0,84702.0,171318.0,259860.0,350340.0,353820.0,
                                268338.0,180882.0,91440.0,66867.0,135210.0,205038.0,276360.0,279042.0,211572.0,142581.0,72060.0,46845.0,94701.0,143574.0,193470.0,195306.0,148047.0,99747.0,50400.0,24576.0,49671.0,75288.0,101430.0,102372.0,77583.0,
                                52260.0,26400.0,22095.0,44688.0,67782.0,91380.0,92178.0,69906.0,47121.0,23820.0,46377.0,93777.0,142206.0,191670.0,193314.0,146571.0,98775.0,49920.0,72906.0,147387.0,223452.0,301110.0,303648.0,230175.0,155082.0,
                                78360.0,101742.0,205638.0,311700.0,419940.0,423420.0,320898.0,216162.0,109200.0,106002.0,214218.0,324660.0,437340.0,440820.0,334038.0,224982.0,113640.0,83292.0,168285.0,254988.0,343410.0,346092.0,262197.0,176556.0,
                                89160.0,58095.0,117351.0,177774.0,239370.0,241206.0,182697.0,122997.0,62100.0,30351.0,61296.0,92838.0,124980.0,125922.0,95358.0,64185.0,32400.0,26970.0,54513.0,82632.0,111330.0,112128.0,84981.0,57246.0,28920.0,56427.0,114027.0,172806.0,232770.0,234414.0,177621.0,119625.0,60420.0,88431.0,178662.0,270702.0,364560.0,367098.0,278100.0,187257.0,94560.0,123042.0,248538.0,376500.0,506940.0,510420.0,386598.0,260262.0,131400.0,127302.0,257118.0,389460.0,524340.0,527820.0,399738.0,269082.0,135840.0,99717.0,201360.0,304938.0,410460.0,413142.0,312822.0,210531.0,106260.0,69345.0,140001.0,211974.0,285270.0,287106.0,217347.0,146247.0,73800.0,36126.0,72921.0,110388.0,148530.0,149472.0,113133.0,76110.0,38400.0}, nd4j::DataType::FLOAT32);

    input.linspace(1);
    weights.linspace(1);
    weights.permutei({2,3,1,0});

    nd4j::ops::deconv2d op;
    auto result = op.execute({&input, &weights}, {}, {5, 5, 1, 1, 0, 0, 1, 1, 0, 0});

    auto z = result->at(0);
    // z->printShapeInfo();
    // z->printBuffer();

    ASSERT_TRUE(exp.isSameShape(z));
    ASSERT_TRUE(exp.equalsTo(z));

    delete result;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, deconv2d_test5) {
    Nd4jLong _expS[] = {4, 2, 3, 8, 8, 192, 64, 8, 1, 16384, 1, 99};
    double _expB[] = {6276.0,12831.0,19668.0,26790.0,27012.0,20703.0,14100.0,7200.0,13719.0,28023.0,42918.0,58410.0,58902.0,45105.0,30693.0,15660.0,22389.0,45696.0,69930.0,95100.0,95910.0,73386.0,49899.0,25440.0,32346.0,65970.0,100884.0,137100.0,138276.0,105726.0,71838.0,36600.0,33726.0,68790.0,105204.0,142980.0,144156.0,110226.0,74898.0,38160.0,27555.0,56154.0,85806.0,116520.0,117474.0,89748.0,60933.0,31020.0,19917.0,40557.0,61926.0,84030.0,84714.0,64671.0,43875.0,22320.0,10752.0,21879.0,33384.0,45270.0,45636.0,34815.0,23604.0,12000.0,7551.0,15456.0,23718.0,32340.0,32562.0,24978.0,17025.0,8700.0,16569.0,33873.0,51918.0,70710.0,71202.0,54555.0,37143.0,18960.0,27114.0,55371.0,84780.0,115350.0,116160.0,88911.0,60474.0,30840.0,39246.0,80070.0,122484.0,166500.0,167676.0,128226.0,87138.0,44400.0,40626.0,82890.0,126804.0,172380.0,173556.0,132726.0,90198.0,45960.0,33180.0,67629.0,103356.0,140370.0,141324.0,107973.0,73308.0,37320.0,23967.0,48807.0,74526.0,101130.0,101814.0,77721.0,52725.0,26820.0,12927.0,26304.0,40134.0,54420.0,54786.0,41790.0,28329.0,14400.0,8826.0,18081.0,27768.0,37890.0,38112.0,29253.0,19950.0,10200.0,19419.0,39723.0,60918.0,83010.0,83502.0,64005.0,43593.0,22260.0,31839.0,65046.0,99630.0,135600.0,136410.0,104436.0,71049.0,36240.0,46146.0,94170.0,144084.0,195900.0,197076.0,150726.0,102438.0,52200.0,47526.0,96990.0,148404.0,201780.0,202956.0,155226.0,105498.0,53760.0,38805.0,79104.0,120906.0,164220.0,165174.0,126198.0,85683.0,43620.0,28017.0,57057.0,87126.0,118230.0,118914.0,90771.0,61575.0,31320.0,15102.0,30729.0,46884.0,63570.0,63936.0,48765.0,33054.0,16800.0,17220.0,34863.0,52932.0,71430.0,72228.0,54831.0,36996.0,18720.0,36327.0,73527.0,111606.0,150570.0,152214.0,115521.0,77925.0,39420.0,57381.0,116112.0,176202.0,237660.0,240198.0,182250.0,122907.0,62160.0,80442.0,162738.0,246900.0,332940.0,336420.0,255198.0,172062.0,87000.0,84702.0,171318.0,259860.0,350340.0,353820.0,268338.0,180882.0,91440.0,66867.0,135210.0,205038.0,276360.0,279042.0,211572.0,142581.0,72060.0,46845.0,94701.0,143574.0,193470.0,195306.0,148047.0,99747.0,50400.0,24576.0,49671.0,75288.0,101430.0,102372.0,77583.0,52260.0,26400.0,22095.0,44688.0,67782.0,91380.0,92178.0,69906.0,47121.0,23820.0,46377.0,93777.0,142206.0,191670.0,193314.0,146571.0,98775.0,49920.0,72906.0,147387.0,223452.0,301110.0,303648.0,230175.0,155082.0,78360.0,101742.0,205638.0,311700.0,419940.0,423420.0,320898.0,216162.0,109200.0,106002.0,214218.0,324660.0,437340.0,440820.0,334038.0,224982.0,113640.0,83292.0,168285.0,254988.0,343410.0,346092.0,262197.0,176556.0,89160.0,58095.0,117351.0,177774.0,239370.0,241206.0,182697.0,122997.0,62100.0,30351.0,61296.0,92838.0,124980.0,125922.0,95358.0,64185.0,32400.0,26970.0,54513.0,82632.0,111330.0,112128.0,84981.0,57246.0,28920.0,56427.0,114027.0,172806.0,232770.0,234414.0,177621.0,119625.0,60420.0,88431.0,178662.0,270702.0,364560.0,367098.0,278100.0,187257.0,94560.0,123042.0,248538.0,376500.0,506940.0,510420.0,386598.0,260262.0,131400.0,127302.0,257118.0,389460.0,524340.0,527820.0,399738.0,269082.0,135840.0,99717.0,201360.0,304938.0,410460.0,413142.0,312822.0,210531.0,106260.0,69345.0,140001.0,211974.0,285270.0,287106.0,217347.0,146247.0,73800.0,36126.0,72921.0,110388.0,148530.0,149472.0,113133.0,76110.0,38400.0,};
    NDArray exp(_expB, _expS);

    auto input = NDArrayFactory::create<double>('c', {2, 3, 4, 4});
    auto weights = NDArrayFactory::create<double>('c', {3, 3, 5, 5});
    auto z = NDArrayFactory::create<double>('c', {2, 3, 8, 8});

    input.linspace(1);
    weights.linspace(1);
    weights.permutei({2,3,1,0});

    nd4j::ops::deconv2d op;
    auto result = op.execute({&input, &weights}, {&z}, {}, {5, 5, 1, 1, 0, 0, 1, 1, 0, 0},{});

    ASSERT_EQ(ND4J_STATUS_OK, result);

    ASSERT_TRUE(exp.isSameShape(&z));
    ASSERT_TRUE(exp.equalsTo(&z));
}

TYPED_TEST(TypedConvolutionTests1, deconv2d_test6) {

    int bS=2, iH=4,iW=4,  iC=3,oC=3,  kH=5,kW=5,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=8,oW=8;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    auto input = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto weights = NDArrayFactory::create<TypeParam>('c', {kH, kW, oC, iC}, {1.f, 76.f, 151.f, 26.f, 101.f, 176.f, 51.f, 126.f, 201.f, 2.f, 77.f, 152.f, 27.f, 102.f, 177.f, 52.f, 127.f, 202.f, 3.f, 78.f, 153.f, 28.f, 103.f, 178.f, 53.f, 128.f, 203.f,
                                    4.f, 79.f, 154.f, 29.f, 104.f, 179.f, 54.f, 129.f, 204.f, 5.f, 80.f, 155.f, 30.f, 105.f, 180.f, 55.f, 130.f, 205.f, 6.f, 81.f, 156.f, 31.f, 106.f, 181.f, 56.f, 131.f, 206.f,
                                    7.f, 82.f, 157.f, 32.f, 107.f, 182.f, 57.f, 132.f, 207.f, 8.f, 83.f, 158.f, 33.f, 108.f, 183.f, 58.f, 133.f, 208.f, 9.f, 84.f, 159.f, 34.f, 109.f, 184.f, 59.f, 134.f, 209.f,
                                    10.f, 85.f, 160.f, 35.f, 110.f, 185.f, 60.f, 135.f, 210.f, 11.f, 86.f, 161.f, 36.f, 111.f, 186.f, 61.f, 136.f, 211.f, 12.f, 87.f, 162.f, 37.f, 112.f, 187.f, 62.f, 137.f, 212.f,
                                    13.f, 88.f, 163.f, 38.f, 113.f, 188.f, 63.f, 138.f, 213.f, 14.f, 89.f, 164.f, 39.f, 114.f, 189.f, 64.f, 139.f, 214.f, 15.f, 90.f, 165.f, 40.f, 115.f, 190.f, 65.f, 140.f, 215.f,
                                    16.f, 91.f, 166.f, 41.f, 116.f, 191.f, 66.f, 141.f, 216.f, 17.f, 92.f, 167.f, 42.f, 117.f, 192.f, 67.f, 142.f, 217.f, 18.f, 93.f, 168.f, 43.f, 118.f, 193.f, 68.f, 143.f, 218.f,
                                    19.f, 94.f, 169.f, 44.f, 119.f, 194.f, 69.f, 144.f, 219.f, 20.f, 95.f, 170.f, 45.f, 120.f, 195.f, 70.f, 145.f, 220.f, 21.f, 96.f, 171.f, 46.f, 121.f, 196.f, 71.f, 146.f, 221.f,
                                    22.f, 97.f, 172.f, 47.f, 122.f, 197.f, 72.f, 147.f, 222.f, 23.f, 98.f, 173.f, 48.f, 123.f, 198.f, 73.f, 148.f, 223.f, 24.f, 99.f, 174.f, 49.f, 124.f, 199.f, 74.f, 149.f, 224.f,
                                    25.f, 100.f, 175.f,50.f, 125.f, 200.f,75.f, 150.f, 225.f});

    auto exp = NDArrayFactory::create<TypeParam>('c', {bS, oC, oH, oW}, {6276.0f,   12831.0f,   19668.0f,   26790.0f,   27012.0f,   20703.0f,   14100.0f,    7200.0f,    13719.0f,   28023.0f,   42918.0f,   58410.0f,   58902.0f,   45105.0f,   30693.0f,   15660.0f,    22389.0f,   45696.0f,   69930.0f,   95100.0f,   95910.0f,   73386.0f,   49899.0f,   25440.0f,    32346.0f,   65970.0f,  100884.0f,  137100.0f,  138276.0f,  105726.0f,   71838.0f,   36600.0f,    33726.0f,   68790.0f,  105204.0f,  142980.0f,  144156.0f,  110226.0f,   74898.0f,   38160.0f,    27555.0f,   56154.0f,   85806.0f,  116520.0f,  117474.0f,   89748.0f,   60933.0f,   31020.0f,    19917.0f,   40557.0f,   61926.0f,   84030.0f,   84714.0f,   64671.0f,   43875.0f,   22320.0f,    10752.0f,   21879.0f,   33384.0f,   45270.0f,   45636.0f,   34815.0f,   23604.0f,   12000.0f,    7551.0f,   15456.0f,   23718.0f,   32340.0f,   32562.0f,   24978.0f,   17025.0f,    8700.0f,    16569.0f,   33873.0f,   51918.0f,   70710.0f,   71202.0f,   54555.0f,   37143.0f,   18960.0f,    27114.0f,   55371.0f,   84780.0f,  115350.0f,  116160.0f,   88911.0f,   60474.0f,   30840.0f,    39246.0f,   80070.0f,  122484.0f,  166500.0f,  167676.0f,  128226.0f,   87138.0f,   44400.0f,    40626.0f,   82890.0f,  126804.0f,  172380.0f,  173556.0f,  132726.0f,   90198.0f,   45960.0f,    33180.0f,   67629.0f,  103356.0f,  140370.0f,  141324.0f,  107973.0f,   73308.0f,   37320.0f,    23967.0f,   48807.0f,   74526.0f,  101130.0f,  101814.0f,   77721.0f,   52725.0f,   26820.0f,    12927.0f,   26304.0f,   40134.0f,   54420.0f,   54786.0f,   41790.0f,   28329.0f,   14400.0f,    8826.0f,   18081.0f,   27768.0f,   37890.0f,   38112.0f,   29253.0f,   19950.0f,   10200.0f,    19419.0f,   39723.0f,   60918.0f,   83010.0f,   83502.0f,   64005.0f,   43593.0f,   22260.0f,    31839.0f,   65046.0f,   99630.0f,  135600.0f,  136410.0f,  104436.0f,   71049.0f,   36240.0f,    46146.0f,   94170.0f,  144084.0f,  195900.0f,  197076.0f,  150726.0f,  102438.0f,   52200.0f,    47526.0f,   96990.0f,  148404.0f,  201780.0f,  202956.0f,  155226.0f,  105498.0f,   53760.0f,    38805.0f,   79104.0f,  120906.0f,  164220.0f,  165174.0f,  126198.0f,   85683.0f,   43620.0f,    28017.0f,   57057.0f,   87126.0f,  118230.0f,  118914.0f,   90771.0f,   61575.0f,   31320.0f,    15102.0f,   30729.0f,   46884.0f,   63570.0f,   63936.0f,   48765.0f,   33054.0f,   16800.0f,    17220.0f,   34863.0f,   52932.0f,   71430.0f,   72228.0f,   54831.0f,   36996.0f,   18720.0f,    36327.0f,   73527.0f,  111606.0f,  150570.0f,  152214.0f,  115521.0f,   77925.0f,   39420.0f,    57381.0f,  116112.0f,  176202.0f,  237660.0f,  240198.0f,  182250.0f,  122907.0f,   62160.0f,    80442.0f,  162738.0f,  246900.0f,  332940.0f,  336420.0f,  255198.0f,  172062.0f,   87000.0f,    84702.0f,  171318.0f,  259860.0f,  350340.0f,  353820.0f,  268338.0f,  180882.0f,   91440.0f,    66867.0f,  135210.0f,  205038.0f,  276360.0f,  279042.0f,  211572.0f,  142581.0f,   72060.0f,    46845.0f,   94701.0f,  143574.0f,  193470.0f,  195306.0f,  148047.0f,   99747.0f,   50400.0f,    24576.0f,   49671.0f,   75288.0f,  101430.0f,  102372.0f,   77583.0f,   52260.0f,   26400.0f,    22095.0f,   44688.0f,   67782.0f,   91380.0f,   92178.0f,   69906.0f,   47121.0f,   23820.0f,    46377.0f,   93777.0f,  142206.0f,  191670.0f,  193314.0f,  146571.0f,   98775.0f,   49920.0f,    72906.0f,  147387.0f,  223452.0f,  301110.0f,  303648.0f,  230175.0f,  155082.0f,   78360.0f,    101742.0f,  205638.0f,  311700.0f,  419940.0f,  423420.0f,  320898.0f,  216162.0f,  109200.0f,    106002.0f,  214218.0f,  324660.0f,  437340.0f,  440820.0f,  334038.0f,  224982.0f,  113640.0f,    83292.0f,  168285.0f,  254988.0f,  343410.0f,  346092.0f,  262197.0f,  176556.0f,   89160.0f,    58095.0f,  117351.0f,  177774.0f,  239370.0f,  241206.0f,  182697.0f,  122997.0f,   62100.0f,    30351.0f,   61296.0f,   92838.0f,  124980.0f,  125922.0f,   95358.0f,   64185.0f,   32400.0f,    26970.0f,   54513.0f,   82632.0f,  111330.0f,  112128.0f,   84981.0f,   57246.0f,   28920.0f,    56427.0f,  114027.0f,  172806.0f,  232770.0f, 

    input.linspace(1);

    nd4j::ops::deconv2d op;
    auto results = op.execute({&input, &weights}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});

    ASSERT_EQ(Status::OK(), results->status());

    auto output = results->at(0);

    ASSERT_TRUE(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete results;
}

TEST_F(ConvolutionTests1, deconv2d_test7) {

    NDArray exp('c', {3, 2, 4, 4}, {218., 227., 236., 245., 254., 263., 272., 281., 290., 299.,  308., 317., 326., 335., 344., 353., 270., 282., 294., 306.,  318., 330., 342., 354., 366., 378., 390., 402., 414., 426.,  438., 450., 650., 659., 668., 677., 686., 695., 704., 713.,  722., 731., 740., 749., 758., 767., 776., 785., 846., 858.,  870., 882., 894., 906., 918., 930., 942., 954., 966., 978.,  990., 1002., 1014., 1026., 1082., 1091., 1100., 1109., 1118., 1127.,  1136., 1145., 1154., 1163., 1172., 1181., 1190., 1199., 1208., 1217.,  1422., 1434., 1446., 1458., 1470., 1482., 1494., 1506., 1518., 1530.,  1542., 1554., 1566., 1578., 1590., 1602.});

    auto input = NDArrayFactory::create<double>('c', {3, 3, 4, 4});
    auto weights = NDArrayFactory::create<double>('c',{1, 1, 2, 3}, {1,3,5,2,4,6});
    auto bias = NDArrayFactory::create<double>('c', {2});

    input.linspace(1);
    bias.linspace(1);

    nd4j::ops::deconv2d op;

    auto result = op.execute({&input, &weights, &bias}, {}, {1, 1, 1, 1, 0, 0, 1, 1, 1, 0});

    ASSERT_EQ(ND4J_STATUS_OK, result->status());

    auto output = result->at(0);

    ASSERT_TRUE(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete result;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests1, deconv2d_test8) {

    int bS=1, iH=7,iW=7,  iC=3,oC=2,  kH=1,kW=1,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=7,oW=7;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iC, iH, iW}, {0.679350, 0.355087, 0.842789, 0.200313, 0.701499, 0.310693, 0.447940, 0.938010, 0.326674, 0.151873, 0.383318, 0.782123, 0.198807,
        0.798564, 0.163263, 0.146968, 0.260897, 0.135058, 0.756209, 0.275454, 0.369088, 0.092826, 0.836492, 0.268413, 0.095062, 0.312795, 0.135918, 0.517544, 0.328703,
        0.061736, 0.396431, 0.248016, 0.548959, 0.115046, 0.814362, 0.721564, 0.404494, 0.299089, 0.403884, 0.988311, 0.022296, 0.927782, 0.318416, 0.068546, 0.284533,
        0.232720, 0.352142, 0.058909, 0.711221, 0.674457, 0.196946, 0.699497, 0.074322, 0.420425, 0.584263, 0.149574, 0.446406, 0.723072, 0.064481, 0.483078, 0.875996,
        0.569819, 0.445863, 0.527755, 0.016646, 0.753678, 0.140636, 0.754129, 0.161932, 0.775037, 0.332645, 0.117394, 0.017711, 0.608476, 0.525152, 0.917194, 0.849891,
        0.589423, 0.852278, 0.390636, 0.889683, 0.669445, 0.698873, 0.961480, 0.157401, 0.157364, 0.493520, 0.569937, 0.126832, 0.115728, 0.786368, 0.737939, 0.490079,
        0.608414, 0.956500, 0.390098, 0.147305, 0.850645, 0.497650, 0.071866, 0.082150, 0.035314, 0.732041, 0.369934, 0.840666, 0.273894, 0.431796, 0.133231, 0.192975,
        0.246897, 0.386418, 0.511541, 0.199036, 0.141631, 0.697699, 0.253631, 0.782218, 0.930099, 0.335512, 0.558808, 0.664358, 0.018851, 0.637559, 0.290430, 0.434902,
        0.842513, 0.466098, 0.381395, 0.523185, 0.990183, 0.925768, 0.643459, 0.016828, 0.918756, 0.228979, 0.006314, 0.665975, 0.190361, 0.595521, 0.698881, 0.221469,
        0.912434, 0.870822, 0.727369, 0.523972, 0.662884, 0.218841});

    NDArray weights('c', {kH, kW, oC, iC}, {0.4195024073123932, 0.22738978266716003, 0.10093523561954498, 0.25008103251457214, 0.3183899223804474, 0.5976081490516663});
    NDArray bias('c', {1, oC}, {0.3596062958240509, 0.6866418123245239});

    NDArray exp('c', {bS, oC, oH, oW}, {0.848190, 0.560603, 0.880509, 0.464103, 0.823376, 0.660138, 0.666382, 0.882257, 0.704650, 0.451427, 0.649734, 0.911822, 0.611581,
        0.847623, 0.568191, 0.439341, 0.710854, 0.473843, 0.927273, 0.605861, 0.724540, 0.530591, 0.804268, 0.478136, 0.602198, 0.639553, 0.669082, 0.855013, 0.678572,
        0.617800, 0.667545, 0.765899, 0.835564, 0.631733, 0.921562, 0.790830, 0.588187, 0.597934, 0.725855, 0.822259, 0.455384, 0.998167, 0.683336, 0.591897, 0.705213,
        0.748148, 0.648922, 0.484723, 0.873482, 1.368675, 0.881096, 1.169214, 0.781504, 1.433406, 1.171439, 1.348675, 1.227033, 1.256600, 0.824772, 1.051633, 1.308692,
        1.148711, 1.334007, 1.014448, 0.813336, 1.408801, 0.916766, 1.583323, 1.362920, 1.226212, 1.149715, 1.330235, 0.770671, 1.285158, 1.105632, 1.272558, 1.590159,
        1.235054, 1.201363, 1.222816, 1.623673, 1.590317, 1.322463, 1.206481, 1.466262, 0.974741, 0.922343, 1.367100, 1.087943, 1.084952, 1.586691, 1.133576, 1.405098,
        1.471922, 1.484062, 1.212039, 1.144419, 1.266123});

    nd4j::ops::deconv2d op;
    auto results = op.execute({&input, &weights, &bias}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});

    ASSERT_EQ(Status::OK(), results->status());

    auto output = results->at(0);

    ASSERT_TRUE(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests1, deconv2d_tf_test1) {

    int bS=2, iH=4,iW=4,  iC=5,oC=10,  kH=2,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=3,oW=3;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, oC});
    auto weights  = NDArrayFactory::create<TypeParam>('c', {kH, kW, iC, oC});
    auto outShape = NDArrayFactory::create<TypeParam>('c', {4}, {static_cast<TypeParam>(bS), static_cast<TypeParam>(iH), static_cast<TypeParam>(iW), static_cast<TypeParam>(iC)});
    auto exp = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC}, {  2.75f,    7.75f,   12.75f,   17.75f,   22.75f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f,  27.75f,   32.75f,   37.75f,   42.75f,   47.75f, 
                                                  55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 105.5f,  115.5f,  125.5f,  135.5f,  145.5f, 
                                                  55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 105.5f,  115.5f,  125.5f,  135.5f,  145.5f, 
                                                  52.75f,   57.75f,   62.75f,   67.75f,   72.75f, 130.5f,  140.5f,  150.5f,  160.5f,  170.5f, 130.5f,  140.5f,  150.5f,  160.5f,  170.5f,  77.75f,   82.75f,   87.75f,   92.75f,   97.75f, 
                                                   2.75f,    7.75f,   12.75f,   17.75f,   22.75f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f,  30.5f,   40.5f,   50.5f,   60.5f,   70.5f,  27.75f,   32.75f,   37.75f,   42.75f,   47.75f, 
                                                  55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 105.5f,  115.5f,  125.5f,  135.5f,  145.5f, 
                                                  55.5f,   65.5f,   75.5f,   85.5f,   95.5f, 161.f,  181.f,  201.f,  221.f,  241.f, 161.f,  181.f,  201.f,  221.f,  241.f, 105.5f,  115.5f,  125.5f,  135.5f,  145.5f, 
                                                  52.75f,   57.75f,   62.75f,   67.75f,   72.75f, 130.5f,  140.5f,  150.5f,  160.5f,  170.5f, 130.5f,  140.5f,  150.5f,  160.5f,  170.5f,  77.75f,   82.75f,   87.75f,   92.75f,   97.75f});
    input = 0.5;
    weights.linspace(0.1, 0.1);

    nd4j::ops::deconv2d_tf op;
    auto results = op.execute({&outShape, &weights, &input}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete results;
}


#endif //LIBND4J_CONVOLUTIONTESTS1_H