cavis/libnd4j/tests_cpu/layers_tests/ConvolutionTests2.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
// @author Yurii Shyrma (iuriish@yahoo.com), created 02.04.2019
//

#ifndef LIBND4J_CONVOLUTIONTESTS2_H
#define LIBND4J_CONVOLUTIONTESTS2_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>

using namespace nd4j;
using namespace nd4j::graph;

class ConvolutionTests2 : public testing::Test {
public:

    const int bS = 2;       // batch size
    const int iD = 1;       // input depth (number of picture channels, for example rgb=3)
    const int iH = 28;      // picture height in pixels
    const int iW = 28;      // picture width in pixels
    const int oD = 3;       // output depth (= N for dense layer)
    const int kH = 5;       // kernel height in pixels
    const int kW = 5;       // kernel width in pixels
    const int sH = 1;       // stride step in horizontal direction
    const int sW = 1;       // stride step in vertical direction
    const int pH = 0;       // padding height
    const int pW = 0;       // padding width
    const int dH = 2;       // dilation height
    const int dW = 2;       // dilation width
    const int oH = (iH - kH - (kH-1)*(dH-1) + 2*pH)/sH + 1;     // output height
    const int oW = (iW - kW - (kW-1)*(dW-1) + 2*pW)/sW + 1;     // output width

};

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, im2col_1) {

    int bS=2, iH=4,iW=3,  iC=4,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int oH = (iH - (kH + (kH-1)*(dH-1)) + 2*pH)/sH + 1; // VALID
    int oW = (iW - (kW + (kW-1)*(dW-1)) + 2*pW)/sW + 1; // VALID

    int paddingMode = 0;             // 1-SAME, 0-VALID;

    NDArray image('c', {bS, iC, iH, iW}, nd4j::DataType::DOUBLE);
    NDArray expected('c', {bS, iC, kH, kW, oH, oW}, {1,  2, 4,  5, 2,  3, 5,  6, 4,  5, 7,  8, 5,  6, 8,  9, 7,  8, 10, 11, 8,  9, 11, 12, 13, 14, 16, 17, 14,
                                                    15, 17, 18, 16, 17, 19, 20, 17, 18, 20, 21, 19, 20, 22, 23, 20, 21, 23, 24, 25, 26, 28, 29, 26, 27, 29, 30,
                                                    28, 29, 31, 32, 29, 30, 32, 33, 31, 32, 34, 35, 32, 33, 35, 36, 37, 38, 40, 41, 38, 39, 41, 42, 40, 41, 43,
                                                    44, 41, 42, 44, 45, 43, 44, 46, 47, 44, 45, 47, 48, 49, 50, 52, 53, 50, 51, 53, 54, 52, 53, 55, 56, 53, 54,
                                                    56, 57, 55, 56, 58, 59, 56, 57, 59, 60, 61, 62, 64, 65, 62, 63, 65, 66, 64, 65, 67, 68, 65, 66, 68, 69, 67,
                                                    68, 70, 71, 68, 69, 71, 72, 73, 74, 76, 77, 74, 75, 77, 78, 76, 77, 79, 80, 77, 78, 80, 81, 79, 80, 82, 83,
                                                    80, 81, 83, 84, 85, 86, 88, 89, 86, 87, 89, 90, 88, 89, 91, 92, 89, 90, 92, 93, 91, 92, 94, 95, 92, 93, 95, 96});

    image.linspace(1, 1);

    nd4j::ops::im2col op;
    auto results = op.execute({&image}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode});
    auto column = results->at(0);

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

    ASSERT_TRUE(expected.isSameShape(column));
    ASSERT_TRUE(expected.equalsTo(column));

    delete results;
}

template <typename T>
class TypedConvolutionTests2 : public testing::Test {
public:

};

typedef ::testing::Types<double, float> TestingTypes;
TYPED_TEST_CASE(TypedConvolutionTests2, TestingTypes);

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, deconv2d_tf_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<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, 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_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;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, Test_DeConv2D_TF_1) {
    auto input0 = NDArrayFactory::create<TypeParam>('c', {4}, {12.f, 5.f, 5.f, 32.f});
    auto input1 = NDArrayFactory::create<TypeParam>('c', {2, 2, 32, 16});
    auto input2 = NDArrayFactory::create<TypeParam>('c', {12, 4, 4, 16});
    auto exp = NDArrayFactory::create<TypeParam>('c', {12, 5, 5, 32});

    nd4j::ops::deconv2d_tf op;
    auto result = op.execute({&input0, &input1, &input2}, {}, {2, 2, 1, 1, 0, 0, 1, 1, 0, 1});
    ASSERT_EQ(Status::OK(), result->status());

    ASSERT_EQ(exp, *result->at(0));

    delete result;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, Test_DeConv2D_TF_2) {
    auto input0 = NDArrayFactory::create<TypeParam>('c', {4}, {3.f, 8.f, 8.f, 16.f});

    auto input1 = NDArrayFactory::create<TypeParam>('c', {7, 7, 16, 5}, {1.05293429f, -0.89349967f, 0.31027254f, 1.22991478f, -0.62926656f, 0.56918693f,
-1.60992694f, 1.10167944f, -0.80843484f, 0.07521993f, -1.15994942f, 0.76016301f, -0.40056285f, -1.16872537f, -0.91384381f, -0.36700436f, 1.82389200f, -1.18200207f, 0.51612782f, -0.92479187f, -0.09307563f, -0.55122334f, 1.23532486f, -1.11124146f, -0.05812126f, 0.68159896f, 0.69125599f, -0.77127314f, -0.10874277f, 0.86469102f,
-1.31614351f, 0.33354419f, -1.71750402f, 0.17197680f, -1.03965557f, 1.10570908f, -1.19115615f, 1.05115080f, 0.18277600f, 1.08820546f, -0.72191417f, -0.10999311f, 1.56521320f, -0.35433730f, -1.11799145f, 0.34499285f, 0.64998639f, -1.64371550f, 0.92592359f, -0.47659501f, 0.49101439f, -0.15613313f, 1.47486567f, 0.43576995f,
2.19538260f, -0.83567709f, -1.21846950f, 0.80400819f, 1.14637423f, -1.01503456f, -0.61992753f, -0.47378838f, 0.86503726f, 0.27147385f, 0.37073180f, -0.19951358f, 0.79167330f, -0.33982825f, 0.18631981f, -1.54715073f, 0.39967480f, 0.95067030f, 1.12508667f, -0.86676019f, -1.10341156f, 2.33141375f, 1.10972047f, 0.71407092f,
1.70640314f, 1.80666339f, 0.59465605f, -0.39653218f, -2.61163163f, -1.15013492f, -1.19908321f, 0.41783467f, -0.22730024f, 0.31425011f, -0.58562893f, -0.10131568f, -0.85047537f, -2.59974790f, 1.22072542f, -2.08812046f, -0.19363593f, -1.27664304f, -0.02703438f, 1.08477545f, -0.65506506f, 0.46040919f, -0.13715318f,
-0.74945593f, -0.69006950f, -1.29617655f, -0.15865716f, 1.38956285f, 0.90216327f, -1.31185400f, -0.15067385f, -0.63093358f, -0.05895613f, 0.26545224f, 0.29332840f, 0.42852548f, 0.72409540f, 0.12879130f, 1.43038857f, 0.68647617f, 2.19654775f, 0.51878077f, -0.03769343f, 0.52877223f, -0.21733910f, 1.13710785f, -0.59003806f,
1.54624867f, -0.64997369f, -1.03239334f, 0.19708300f, 0.68658423f, 0.71048903f, -1.55250466f, -1.38636279f, 0.32385820f, 0.81226677f, 0.19209047f, -0.23002781f, -0.63631231f, 1.02101684f, 0.65428704f, -0.17206922f, 1.09488952f, 1.03022420f, -0.95567745f, -0.07595373f, -1.48606372f, 2.57174873f, -1.75366247f, 1.12913883f,
0.97053039f, -0.28552356f, 0.56511772f, -0.79568213f, 0.07561764f, -1.02085686f, 1.05770981f, -1.25715709f, 0.42046708f, -2.57390857f, 0.96947151f, 1.05215812f, 0.65624017f, -1.29019403f, 0.64157075f, -0.40509227f, -0.65354455f, 0.42348680f, -1.34107757f, 0.05931387f, -0.54337227f, 0.95460182f, 1.59319806f, -0.44433126f,
-0.33717924f, 0.79566282f, 0.50112695f, -0.22244534f, 1.76904583f, -0.89817202f, 1.82985342f, 0.17671813f, 0.80720717f, 1.32469308f, 0.39417782f, -0.23720963f, 0.96796370f, -1.02348757f, -0.86615551f, -1.58120525f, -0.37634999f, 0.00905940f, 0.01880967f, 1.75771821f, -0.64372772f, 0.36687651f, 0.15854552f, -0.67599791f,
0.53726906f, -1.20158446f, -1.78549063f, 0.96476388f, -0.66158366f, -0.41681561f, -0.97541636f, 2.35928202f, 0.32130197f, 1.06886065f, 1.38736427f, -0.73718959f, 0.11215294f, 2.12865782f, -0.37927702f, 0.55621815f, -1.10108411f, -0.02032263f, 0.29595461f, 1.58737493f, 1.24001300f, -0.66748160f, 0.80729002f, -0.10575818f,
-1.03175950f, 1.80755460f, 0.10825710f, 2.20666361f, 1.33633149f, 1.39290452f, 0.45211342f, -0.07837920f, 2.08304930f, -0.28387162f, -0.70775616f, 0.43626297f, 0.53556961f, 0.06201901f, -0.59255266f, -0.11854446f, 2.10024118f, 0.37638292f, -0.56178707f, -0.25220188f, -1.23731256f, -1.30002999f, 0.34283713f, 0.30502397f,
-1.09233856f, 1.12430644f, 0.52273953f, -0.68507338f, -0.69913578f, 0.88440478f, -0.76959240f, 1.07093310f, -0.34802195f, 0.35683727f, -0.76079178f, -1.92807376f, 0.84499562f, 1.39131641f, 0.44825050f, 0.34567752f, 0.44607711f, -1.00986362f, -0.50038189f, -0.09060892f, -2.55645394f, 0.56416476f, -0.83058155f, -0.65931624f,
-0.73649710f, 0.59814465f, -0.86736494f, -0.32200798f, -1.28087902f, -0.76818323f, 0.86848933f, -0.98678392f, -1.30813944f, -0.20255326f, 0.26557815f, -0.31090519f, -1.46331608f, -0.62782109f, 0.59034890f, 1.63147473f, -0.17727259f, -0.37636510f, 1.27368402f, 0.19096918f, -0.29936951f, -1.99038267f, 0.54831523f, 0.48849005f, -2.55680346f, -0.63126534f, 1.21715927f, 1.22841084f, -0.67416084f, 0.02927168f, -0.36693662f, 0.63204330f, 0.13721083f, 0.28742912f, 0.19470036f, 0.74873924f, -1.47602463f, 0.86264688f, -0.23730527f, -0.99978864f, -1.17048764f, -0.34996086f, 1.43019187f, 0.26224539f, 0.60689932f, -0.75002515f, -0.79823422f, -1.37300086f, -0.19951135f, -0.12150808f, -0.75272322f, 0.23755015f, 0.31270382f, 1.66539109f, -1.04104745f, 0.79540199f, -0.54042423f, -0.54150617f, 0.43871084f, 0.24163951f, -0.24517761f, -0.66178995f, -1.13064528f, -0.84426326f, 0.56437236f, 0.09088907f, -0.82823074f, 0.81753862f, -1.74096012f, -1.80599844f, -0.60943592f, 1.36094582f, -1.47762752f, 0.15931177f, 1.05569172f, 0.36751524f, 0.06497604f, 0.13536447f, -1.57156146f, 0.22783801f, -0.96910107f, -1.24294984f, -1.47147155f, -1.04790676f, 0.64629447f, -0.32266054f, -0.55675793f, -0.95612079f, -0.23005411f, -0.75229394f, 0.03050950f, -1.72484553f, -2.06055546f, 0.19892083f, -0.13597751f, 0.65180075f, 0.27096850f, 0.08977254f, 0.57564765f, -0.43227410f, 0.09541437f, -0.00358280f, 0.65680492f, 0.04006556f, 0.57160908f, 0.43821687f, 1.96118212f, 0.42602235f, -0.36731303f, 0.67200917f, -0.56667900f, 0.44014785f, 0.06970236f, -1.34415269f, -1.13301528f, -0.08848868f, 0.35615012f, -0.06426942f, -0.81406075f, 0.94097465f, -0.54560357f, -0.65877116f, -1.29646838f, -1.13109028f, -1.64186084f, -2.12723470f, 1.86027610f, 1.22621441f, 0.26098135f, -0.05608099f, 0.21143445f, -0.87244326f, 0.79408187f, 1.24279130f, 0.14458629f, 0.25532281f, -1.24023473f, 2.42278886f, 0.00405578f, -1.00119174f, 1.19856644f, -1.37395728f, -0.16656208f, 0.46858498f, -0.00678801f, -0.34960639f, 0.16614936f, 2.41560221f, -0.53880709f, 0.91618651f, -1.77009308f, 0.32911557f, 0.30216452f, 0.02881077f, 0.77705866f, 0.27061903f, -0.07440855f, -1.14010465f, 1.25383139f, -1.58615100f, 1.04185510f, 0.15140508f, -0.88059032f, -0.33872122f, -0.42526904f, 2.17365575f, 0.29308075f, -2.24234557f, -1.03164542f, -0.09263755f, 0.08050421f, -0.74946511f, -0.64589006f, -1.13416314f, -0.64989561f, 0.16502371f, -0.33831969f, 0.22832428f, -0.08389475f, -0.28009200f, 1.34536922f, -0.19075738f, 0.36238208f, 0.83690089f, 0.26144615f, 0.04457319f, -2.55585861f, -0.01807522f, 1.68334866f, -0.05795629f, -0.21315987f, -1.84039557f, 0.06512877f, -1.77318645f, -0.27637982f, 0.20439345f, 0.67558700f, -0.77179354f, -0.17902173f, 0.70381826f, -0.40395790f, -0.96492916f, 0.84138173f, 2.43879008f, -0.32297835f, -1.74370265f, -0.10330839f, -1.07465363f, 1.85030377f, -0.59153467f, 0.99667048f, -0.56753993f, 0.57383025f, -1.90630126f, 1.24299097f, 0.22797665f, 0.30468231f, -0.07360230f, 1.64654350f, 0.57195550f, 0.03227921f, 1.11005175f, 0.00088721f, 1.19266295f, 0.61323351f, 0.13754399f, 0.59900171f, -0.75831634f, 1.11500823f, 0.99747783f, -1.36923385f, 1.26563418f, 0.01253266f, 0.35483193f, 1.95143735f, -2.02703261f, -1.38265920f, -0.02404256f, 2.02788448f, -0.75144875f, -0.58445263f, 0.26129767f, 0.60691077f, -1.84661067f, 0.65872228f, -0.58298993f, 0.33067298f, -0.09431327f, 0.43333948f, -1.52616286f, -0.25961858f, -1.65459549f, -0.72950101f, -0.89906919f, -0.80081612f, -1.32189929f, -1.36574399f, -0.35809481f, 0.36385000f, 0.31480747f, -0.35797358f, -1.04066050f, 0.07971872f, -0.21176252f, -0.76559299f, -0.10352154f, 0.29248312f, -1.75030553f, 0.68219930f, 0.56189102f, -1.11212170f, 0.06501702f, -0.07131009f, 1.23410738f, 0.29311740f, -1.02052307f, 1.40220940f, -1.00995779f, 0.57955760f, 0.22640309f, 0.74853230f, -0.02586563f, -0.33427954f, 1.70311153f, -0.53405988f, 0.90975094f, -0.46450076f, 0.19904344f, 0.28559047f, 0.23167793f, -0.69065529f, -0.17176504f, -0.29301846f, -0.85477978f, -0.00267053f, -0.28529504f, -0.64201307f, 1.03479636f, 1.03805065f, 0.83270210f, -0.09405448f, 2.50615931f, 0.62019676f,

    auto input2 = NDArrayFactory::create<TypeParam>('c', {3, 4, 4, 5}, {0.98114507f, 0.96400015f, 0.58669623f, 0.60073098f, 0.75425418f, 0.44258752f, 0.76373084f, 0.96593234f, 0.34067846f, 0.57962620f, 0.77517051f, 0.97472977f, 0.79237527f, 0.68690428f, 0.21719366f, 0.79959206f, 0.84814187f, 0.22496814f, 0.08646965f, 0.31110474f, 0.79813162f, 0.19661444f, 0.57760099f, 0.72138960f, 0.15244268f, 0.87687051f, 0.11130344f, 0.01087698f, 0.34817841f, 0.54992017f, 0.23443850f, 0.31725614f, 0.59755220f, 0.20364695f, 0.00531392f, 0.23403114f, 0.07442912f, 0.83707647f, 0.89291743f, 0.09044587f, 0.69041462f, 0.29904183f, 0.61904680f, 0.85306847f, 0.34467042f, 0.95839152f, 0.54517124f, 0.29640937f, 0.94855959f, 0.95970016f, 0.94045145f, 0.95510301f, 0.34666505f, 0.34717010f, 0.69245678f, 0.71669175f, 0.59043738f, 0.64924132f, 0.06033522f, 0.60185199f, 0.04690073f, 0.59241154f, 0.40229547f, 0.23002481f, 0.45161195f, 0.73743778f, 0.93209113f, 0.37294358f, 0.50177744f, 0.15072501f, 0.26146917f, 0.05252146f, 0.04758931f, 0.76448288f, 0.85149045f, 0.08840467f, 0.07692576f, 0.33180160f, 0.27241259f, 0.74834620f, 0.56453640f, 0.23057286f, 0.68429752f, 0.11961551f, 0.39045977f, 0.44356094f, 0.77018807f, 0.07984410f, 0.47926806f, 0.26165759f, 0.18606064f, 0.89972877f, 0.17962874f, 0.47273120f, 0.64641705f, 0.61890443f, 0.58730015f, 0.25937832f, 0.35231561f, 0.10243882f, 0.17459193f, 0.95906995f, 0.09227025f, 0.30003223f, 0.41601210f, 0.38269713f, 0.84799751f, 0.59295173f, 0.76277990f, 0.68910424f, 0.37672606f, 0.40675461f, 0.94346058f, 0.91438505f, 0.84728183f, 0.64367667f, 0.74899979f, 0.60570691f, 0.16417363f, 0.68852426f, 0.85486889f, 0.22585792f, 0.86953176f, 0.07465519f, 0.93096301f, 0.38008822f, 0.38752587f, 0.44004038f, 0.13170612f, 0.94541045f, 0.89349973f, 0.69245307f, 0.94978877f, 0.98776658f, 0.79445884f, 0.30607409f, 0.58264961f, 0.37980538f, 0.41810784f, 0.48903038f, 0.51615888f, 0.57682794f, 0.82481897f, 0.78341080f, 0.48446465f, 0.17447931f, 0.71125424f, 0.30263851f, 0.70675352f, 0.03215584f, 0.92381065f, 0.22343694f, 0.08851149f, 0.91402490f, 0.70074717f, 0.30912192f, 0.37723206f, 0.97579397f, 0.23554587f, 0.95939133f, 0.41565709f, 0.01741416f, 0.58362787f, 0.22106662f, 0.89065537f, 0.31900249f, 0.41280911f, 0.67947610f, 0.04545590f, 0.15352812f, 0.85412524f, 0.84933222f, 0.80000225f, 0.93147073f, 0.70094105f, 0.69269875f, 0.95282194f, 0.65913582f, 0.79186874f, 0.59855248f, 0.39707430f, 0.95126239f, 0.15618217f, 0.33446689f, 0.98123758f, 0.84770758f, 0.98081012f, 0.54427413f, 0.18728519f, 0.89792955f, 0.53360126f, 0.72812986f, 0.13307744f, 0.51217443f, 0.66708084f, 0.29416915f, 0.31298995f, 0.39155037f, 0.29288291f, 0.87063305f, 0.61759154f, 0.73723332f, 0.37167635f, 0.82122716f, 0.22937430f, 0.76570536f, 0.47911792f, 0.02826214f, 0.94277323f, 0.59945469f, 0.19042060f, 0.68173155f, 0.82771295f, 0.95649538f, 0.40833101f, 0.90838542f, 0.55245881f, 0.49011012f, 0.36773444f, 0.34513527f, 0.42050683f, 0.16113964f, 0.30969388f, 0.27174174f, 0.12117655f, 0.35270175f, 0.81967867f, 0.63723136f, 0.84309389f, 0.71822576f, 0.84883484f, 0.32306117f, 0.08176457f, 0.56175486f, 0.34892198f, 0.09306929f, 0.85437582f, 0.13925577f, 0.48629188f, 0.29923539f});
    auto exp = NDArrayFactory::create<TypeParam>('c', {3, 8, 8, 16}, {5.98743296f, -2.83037376f, -0.87943113f, 1.41339970f, 1.32433391f, -1.20299149f, -0.02893090f, 2.05326009f, 1.19417048f, 5.58212376f, 3.28139353f, 1.19237995f, -1.09431255f, -2.55264497f, 3.11014652f, 6.81296825f, -2.09029293f, -4.32068443f, -0.52808392f, -1.97968531f, -0.18673831f, 0.84605980f, 4.55825520f, 2.71503139f, 0.15210046f, 0.85310984f, -3.82062817f, 2.76470995f, 3.69004202f, -1.45017099f, -2.59361267f, -1.35094655f, 7.24145126f, -5.25432396f, 0.19920218f, -4.30596399f, 1.35318923f, -3.88142037f, 3.67493343f, 2.25931478f, 2.87630725f, 1.66349852f, 6.21347952f, 0.94105923f, -1.61742055f, -2.35699606f, 0.12850338f, 1.79141688f, -2.09535933f, -6.35418081f, -0.06303531f, -4.38615131f, 0.48237842f, 0.26528549f, 3.38231516f, 3.76315165f, -0.40254810f, -0.23716694f, -6.13381910f, -0.41950428f, -0.89680839f, -1.46491277f, -1.98541689f, -0.99357355f, 5.58237648f, -2.38937521f, -0.00872564f, -2.37138414f, 4.91117287f, -4.51916361f, 0.97943687f, 2.91052818f, -2.50362611f, 1.70252812f, 5.04137802f, 3.57108784f, -1.87532270f, -3.66677809f, -2.38861251f, 5.55765152f, -7.27571774f, -1.68887305f, -0.72266489f, -4.42809057f, -0.92118186f, 1.02381468f, 4.44284725f, 5.17150497f, -0.42438728f, 2.02693963f, -1.36484981f, -1.47912180f, 0.26649538f, -0.02091765f, -2.86906910f, -3.03046989f, 1.35122132f, -3.21707630f, 2.21112418f, 0.24121630f, 3.96940088f, -7.66105747f, 2.76352382f, -0.99061489f, -2.16720009f, -1.63170409f, 1.12701774f, -1.02415371f, -0.90435314f, -1.51372027f, -0.76884907f, 0.39066136f, -0.89562428f, -2.03204703f, 1.28074932f, -2.14551091f, -2.36843777f, 0.46580017f, 0.75451565f, -0.00336730f, -1.06597757f, 3.27195978f, -0.41307712f, -0.10376054f, -1.34102952f, -2.22901654f, 2.31929803f, 1.40851438f, -2.23774385f, 0.20417206f, -1.12153268f, -0.13188094f, -3.96649432f, 2.10269976f, 0.49845099f, 6.18937683f, -0.51783508f, -0.48048639f, -1.92970264f, 3.16670656f, 1.13355756f, -0.07890664f, 1.31536257f, -0.43924797f, -0.04562932f, -0.87974954f, 0.75411212f, -2.39745235f, -3.97132111f, 0.37202546f, -2.40399146f, -1.50796390f, -3.08302689f, 0.23075986f, -0.94316757f, 1.34948587f, 0.58591264f, 2.18529797f, 7.97652435f, 2.32798409f, -4.09404373f, 0.89634895f, 0.77697754f, -0.65091681f, -7.05506849f, 5.86194515f, 2.51394033f, 4.69959354f, 0.20835471f, 3.18049693f, -1.29682434f, 3.70832396f, -0.48123091f, -1.67904007f, -1.35418940f, 1.58435583f, -1.13851106f, -1.19225955f, 0.59713769f, -5.80462933f, -7.45143986f, -1.08658695f, 1.03244078f, -1.75307107f, -7.07100582f, 3.85825157f, 1.62127817f, 2.32572675f, 0.56171900f, -0.80591971f, 3.98835945f, 0.15742642f, -2.97832179f, 0.13821673f, -0.72556758f, -0.84936106f, -7.28444147f, 3.94134307f, 0.80779338f, 7.47784615f, 8.23335075f, 4.80595016f, -4.89574575f, 4.03362942f, -6.67522192f, -4.55204487f, 2.12511182f, -2.70781207f, -1.57226098f, -3.08408356f, -0.30812448f, -5.32870674f, -5.13238287f, 0.49605465f, -0.55042171f, 0.46324944f, -3.83545256f, -0.12562510f, -0.20978995f, -0.13068712f, -1.92144060f, -1.68787408f, 5.45581436f, -0.79583496f, -2.38866687f, -3.90546346f, -0.47028148f, -0.14319679f, -3.37016582f, 2.00905991f, -1.21345615f, 1.81376505f, 7.73004007f, 0.74310112f, -4.64536428f, 3.78111577f, -9.05182457f, -0.10674095f, 1.53476238f, 0.63345337f, -0.40907967f, -1.44729769f, -1.87145400f, -2.46623540f, 1.07472968f, 0.77390999f, -3.93438888f, 4.49174690f, -0.96686655f, 1.92278123f, 0.30049133f, -0.02388665f, -1.99777114f, -3.23885751f, 5.87784004f, 2.13776040f, 3.56758308f, -3.37774134f, -3.67526293f, 1.63700044f, -1.69959962f, -0.99112594f, 6.03103638f, 1.67399430f, -1.28699589f, 7.16759014f, 12.63490295f, 3.62937450f, -4.75982571f, 2.17861104f, -2.03065681f, 4.30207729f, -0.46797156f, -2.96022511f, -6.02702332f, 3.09229851f, -1.39771092f, -0.03471333f, 3.22175527f, 5.63565636f, 1.78195477f, -0.63545251f, -3.99497652f, 1.46043062f, 4.60050488f, -2.96651959f, -2.03159475f, -1.52386189f, -0.15129802f, -3.90390921f, -0.63852370f, 0.79210538f, 2.35288715f, -5.55609035f, 5.36427498f, -0.60248077f, -0.26181316f,

    nd4j::ops::deconv2d_tf op;
    auto result = op.execute({&input0, &input1, &input2}, {}, {7,7,  2,2,  0,0,  1,1,  1,1});
    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(ConvolutionTests2, Test_Dilation2D_Again_1) {
    auto x = NDArrayFactory::create<double>('c', {4, 128, 128, 4});
    auto w = NDArrayFactory::create<double>('c', {4, 5, 4});
    auto exp = NDArrayFactory::create<double>('c', {4, 64, 43, 4});


    nd4j::ops::dilation2d op;
    auto result = op.execute({&x, &w}, {}, {1, 1,5,7,1, 1,2,3,1});
    ASSERT_EQ(Status::OK(), result->status());

    auto z = result->at(0);

    ASSERT_TRUE(exp.isSameShape(z));

    delete result;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, Test_Dilation2D_Again_2) {
    auto x = NDArrayFactory::create<double>('c', {4, 26, 19, 4});
    auto w = NDArrayFactory::create<double>('c', {11, 7, 4});

    nd4j::ops::dilation2d op;
    auto result = op.execute({&x, &w}, {}, {0, 1,2,3,1, 1,3,2,1});
    ASSERT_EQ(Status::OK(), result->status());

    delete result;
}

TYPED_TEST(TypedConvolutionTests2, sconv2d_bp_1) {
    TypeParam _expGradWpB[] = {1603.7102981f,  10645.6278024f,   5975.4227995f,  17697.0903052f,    12133.6353024f,  26535.0528052f,   1779.221097f,   11795.5686029f,    6721.9835994f,  19904.0811062f,  13775.2461029f,  30123.0936062f,    1954.7318976f,  12945.5094033f,   7468.5443993f,  22111.071907f,    15416.8569033f,  33711.134407f,    2130.2426974f,  14095.4502038f,    8215.1051992f,  24318.0627081f,  17058.4677038f,  37299.1752081f,    2305.7534972f,  15245.3910042f,   8961.6659991f,  26525.0535091f,    18700.0785042f,  40887.2160091f,   2481.2642970f,  16395.3318047f,    9708.2267991f,  28732.0443100f,  20341.6893047f,  44475.2568100f,    2656.7750968f,  17545.2726051f,  10454.7875990f,  30939.0351110f,    21983.3001051f,  48063.2976110f,   2832.2858966f,  18695.2134056f,    11201.3483989f,  33146.0259119f,  23624.9109056f,  51651.3384119f,    3007.7966964f,  19845.1542060f,  11947.9091988f,  35353.0167129f,    25266.5217060f,  55239.3792129f,   3183.3074962f,  20995.095006f,    12694.4699987f,  37560.007513f,   26908.132506f,   58827.4200139f};
    Nd4jLong _expGradWpS[] {4, 10, 6, 1, 1, 6, 1, 1, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99};
    NDArray expGWP(_expGradWpB, _expGradWpS);
    expGWP.permutei({2,3,1,0});

    TypeParam _expGradWdB[] = {2074.21032f, 2082.76104f, 2091.31176f, 2099.86248f, 2108.4132f, 2159.71752f, 2168.26824f, 2176.81896f, 2185.36968f, 2193.9204f, 2245.22472f, 2253.77544f, 2262.32616f, 2270.87688f, 2279.4276f, 2330.73192f, 2339.28264f, 2347.83336f, 2356.38408f, 2364.9348f, 2416.23912f, 2424.78984f, 2433.34056f, 2441.89128f, 2450.442f, 3112.99344f, 3122.06328f, 3131.13312f, 3140.20296f, 3149.2728f, 3203.69184f, 3212.76168f, 3221.83152f, 3230.90136f, 3239.9712f, 3294.39024f, 3303.46008f, 3312.52992f, 3321.59976f, 3330.6696f, 3385.08864f, 3394.15848f, 3403.22832f, 3412.29816f, 3421.368f, 3475.78704f, 3484.85688f, 3493.92672f, 3502.99656f, 3512.0664f, 4255.60056f, 4265.18952f, 4274.77848f, 4284.36744f, 4293.9564f, 4351.49016f, 4361.07912f, 4370.66808f, 4380.25704f, 4389.846f, 4447.37976f, 4456.96872f, 4466.55768f, 4476.14664f, 4485.7356f, 4543.26936f, 4552.85832f, 4562.44728f, 4572.03624f, 4581.6252f, 4639.15896f, 4648.74792f, 4658.33688f, 4667.92584f, 4677.5148f, 2140.10988f, 2148.92016f, 2157.73044f, 2166.54072f, 2175.351f, 2228.21268f, 2237.02296f, 2245.83324f, 2254.64352f, 2263.4538f, 2316.31548f, 2325.12576f, 2333.93604f, 2342.74632f, 2351.5566f, 2404.41828f, 2413.22856f, 2422.03884f, 2430.84912f, 2439.6594f, 2492.52108f, 2501.33136f, 2510.14164f, 2518.95192f, 2527.7622f, 3204.849f, 3214.1784f, 3223.5078f, 3232.8372f, 3242.1666f, 3298.143f, 3307.4724f, 3316.8018f, 3326.1312f, 3335.4606f, 3391.437f, 3400.7664f, 3410.0958f, 3419.4252f, 3428.7546f, 3484.731f, 3494.0604f, 3503.3898f, 3512.7192f, 3522.0486f, 3578.025f, 3587.3544f, 3596.6838f, 3606.0132f, 3615.3426f, 4373.41212f, 4383.26064f, 4393.10916f, 4402.95768f, 4412.8062f, 4471.89732f, 4481.74584f, 4491.59436f, 4501.44288f, 4511.2914f, 4570.38252f, 4580.23104f, 4590.07956f, 4599.92808f, 4609.7766f, 4668.86772f, 4678.71624f, 4688.56476f, 4698.41328f, 4708.2618f, 4767.35292f, 4777.20144f, 4787.04996f, 4796.89848f, 4806.747f};
    Nd4jLong _expGradWdS[] = {4, 2, 3, 5, 5, 75, 25, 5, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99};
    NDArray expGWD(_expGradWdB, _expGradWdS);
    expGWD.permutei({2,3,1,0});

    TypeParam _expEB[] = {5.0103f, 10.17147f, 15.48408f, 20.9487f, 26.5659f, 26.6832f, 21.65628f, 16.47507f, 11.139f, 5.6475f, 10.79727f, 21.90255f, 33.31698f, 45.0417f, 57.07785f, 57.3267f, 46.49334f, 35.34513f, 23.88093f, 12.0996f, 17.37801f, 35.22744f, 53.55f, 72.3474f, 91.62135f, 92.016f, 74.57958f, 56.66148f, 38.25999f, 19.3734f, 24.76962f, 50.18034f, 76.23444f, 102.9342f, 130.2819f, 130.8366f, 105.9834f, 80.47542f, 54.31038f, 27.486f, 32.9892f, 66.79545f, 101.4216f, 136.8705f, 173.145f, 173.874f, 140.7732f, 106.83825f, 72.0663f, 36.4545f, 33.8298f, 68.49375f, 103.9947f, 140.3355f, 177.519f, 178.248f, 144.3066f, 109.51395f, 73.8672f, 37.3635f, 28.85658f, 58.39302f, 88.6116f, 119.5146f, 151.1043f, 151.716f, 122.76444f, 93.11934f, 62.77842f, 31.7394f, 23.00409f, 46.52748f, 70.57188f, 95.139f, 120.23055f, 120.7107f, 97.6311f, 74.02194f, 49.88151f, 25.2081f, 16.25523f, 32.86293f, 49.82424f, 67.1403f, 84.81225f, 85.1466f, 68.83818f, 52.17045f, 35.14227f, 17.7525f, 8.5929f, 17.36517f, 26.31738f, 35.4501f, 44.7639f, 44.9382f, 36.31728f, 27.51357f, 18.5265f, 9.3555f, 8.63807f, 17.45032f, 26.43736f, 35.5998f, 44.93825f, 45.1399f, 36.46882f, 27.6199f, 18.59253f, 9.3861f, 18.18615f, 36.72737f, 55.62488f, 74.8799f, 94.49365f, 94.9122f, 76.65698f, 58.03937f, 39.05815f, 19.7121f, 28.66254f, 57.86775f, 87.61746f, 117.9135f, 148.7577f, 149.4084f, 120.63768f, 91.31331f, 61.43346f, 30.9963f, 40.08554f, 80.90806f, 122.47f, 164.7738f, 207.8219f, 208.72f, 168.48412f, 127.49662f, 85.75506f, 43.257f, 52.47345f, 105.8849f, 160.2374f, 215.534f, 271.77775f, 272.9385f, 220.2695f, 166.6442f, 112.05955f, 56.5125f, 53.82975f, 108.6158f, 164.3612f, 221.069f, 278.74225f, 279.903f, 225.8777f, 170.8778f, 114.90025f, 57.942f, 45.14002f, 91.0585f, 137.75788f, 185.2406f, 233.5091f, 234.4682f, 189.16564f, 143.06998f, 96.17878f, 48.4896f, 35.43048f, 71.45487f, 108.075f, 145.2927f, 183.1098f, 183.852f, 148.29504f, 112.13319f, 75.36462f, 37.9875f, 24.68283f, 49.76831f, 75.25766f, 101.1521f, 127.45285f, 127.9629f, 103.1927f, 78.01253f, 52.42117f, 26.4174f, 12.87877f, 25.96222f, 39.25096f, 52.7456f, 66.44675f, 66.7094f, 53.78542f, 40.6531f, 27.31183f, 13.761f, 12.59184f, 25.38317f, 38.37464f, 51.5669f, 64.9606f, 65.2566f, 52.61336f, 39.76673f, 26.71606f, 13.4607f, 26.23903f, 52.88419f, 79.93678f, 107.3981f, 135.26945f, 135.8777f, 109.53262f, 82.77361f, 55.59937f, 28.0086f, 40.96107f, 82.54206f, 124.74492f, 167.5716f, 211.02405f, 211.9608f, 170.83578f, 129.07914f, 86.68893f, 43.6632f, 56.77746f, 114.39578f, 172.85756f, 232.1654f, 292.3219f, 293.6034f, 236.60084f, 178.74182f, 120.02374f, 60.444f, 73.7077f, 148.48435f, 224.3332f, 301.2575f, 379.2605f, 380.903f, 306.9058f, 231.82015f, 155.6428f, 78.3705f, 75.6397f, 152.36785f, 230.1877f, 309.1025f, 389.1155f, 390.758f, 314.8288f, 237.79165f, 159.6433f, 80.3805f, 62.89546f, 126.67598f, 191.34416f, 256.9026f, 323.3539f, 324.7004f, 261.56684f, 197.53262f, 132.59514f, 66.7518f, 48.97887f, 98.63226f, 148.96212f, 199.9704f, 251.65905f, 252.6933f, 203.53098f, 153.68244f, 103.14573f, 51.9189f, 33.87043f, 68.19769f, 102.98308f, 138.2279f, 173.93345f, 174.6392f, 140.64322f, 106.18261f, 71.25607f, 35.8623f, 17.55064f, 35.33327f, 53.34854f, 71.5971f, 90.0796f, 90.4406f, 72.82556f, 54.97463f, 36.88716f, 18.5625f, 13.0455f, 26.44707f, 40.20528f, 54.3207f, 68.7939f, 68.9112f, 55.84908f, 42.42747f, 28.6458f, 14.5035f, 27.89367f, 56.50575f, 85.83738f, 115.8897f, 146.66385f, 146.9127f, 118.98294f, 90.32793f, 60.94653f, 30.8376f, 44.56161f, 90.21024f, 136.9476f, 184.7754f, 233.69535f, 234.09f, 189.46998f, 143.75268f, 96.93639f, 49.0194f, 63.06642f, 127.59474f, 193.58724f, 261.0462f, 329.9739f, 330.5286f, 267.3786f, 202.75302f, 136.64958f, 69.066f, 83.4252f, 168.69345f, 255.8076f, 344.7705f, 435.585f, 436.314f, 352.7772f, 267.38025f, 180.1203f, 90.9945f, 84.2658f, 170.39175f, 258.3807f, 348.2355f, 439.959f, 440.688f, 356.3106f, 270.05595f, 181.9212f, 91.9035f, 71.25738f, 144.01542f, 218.2764f, 294.0426f, 371.3163f, 371.928f, 300.57564f, 227.70894f, 153.32562f, 77.4234f, 56.34369f, 113.82228f, 172.43748f, 232.191f, 293.08455f, 293.
    Nd4jLong _expES[] = {4, 2, 3, 10, 10, 300, 100, 10, 1, typeid(TypeParam) == typeid(float) ? 8192 : 16384, 1, 99};
    NDArray expE(_expEB, _expES);

    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});

    auto epsilon = NDArrayFactory::create<TypeParam>('c', {2, 3, 10, 10});
    auto epsilonNext = NDArrayFactory::create<TypeParam>('c', {2, 10, 6, 6});

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

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

    nd4j::ops::sconv2d_bp op;
    auto resultBP = op.execute({&input, &epsilonNext, &weightsD, &weightsP },{}, {5, 5, 1, 1, 0, 0, 1, 1, 0}, {});

    ASSERT_EQ(3, resultBP->size());

    auto _epsilon = resultBP->at(0);
    auto _gradWD = resultBP->at(1);
    auto _gradWP = resultBP->at(2);

    //_gradWP->printBuffer("gradWP");

    ASSERT_TRUE(_gradWP->isSameShape(&expGWP));
    ASSERT_TRUE(_gradWP->isSameShape(&weightsP));

    ASSERT_TRUE(_gradWP->equalsTo(&expGWP));

    //_gradWD->printShapeInfo("gradWD shape");

    ASSERT_TRUE(_gradWD->isSameShape(&expGWD));
    ASSERT_TRUE(_gradWD->isSameShape(&weightsD));
// _gradWD->printIndexedBuffer();
    ASSERT_TRUE(_gradWD->equalsTo(&expGWD));

    ASSERT_TRUE(_epsilon->isSameShape(&input));
    ASSERT_TRUE(_epsilon->isSameShape(&expE));

    ASSERT_TRUE(_epsilon->equalsTo(&expE));

    delete resultBP;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, sconv2d_bp_2) {

    int bS=3, iH=16,iW=16,  iC=3,mC=3,  kH=1,kW=1,  sH=1,sW=1,  pH=0,pW=0,  dH=2,dW=2;
    int       oH=16,oW=16;
    int       oC=2;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iC, iH, iW}, typeid(TypeParam) == typeid(float) ? nd4j::DataType::FLOAT32 : nd4j::DataType::DOUBLE);
    NDArray gradO('c', {bS, oC, oH, oW}, typeid(TypeParam) == typeid(float) ? nd4j::DataType::FLOAT32 : nd4j::DataType::DOUBLE);
    NDArray weightsDepth('c', {kH, kW, iC, mC}, typeid(TypeParam) == typeid(float) ? nd4j::DataType::FLOAT32 : nd4j::DataType::DOUBLE);
    NDArray weightsPoint('f', {1, 1, iC*mC, oC}, typeid(TypeParam) == typeid(float) ? nd4j::DataType::FLOAT32 : nd4j::DataType::DOUBLE);
    NDArray bias('c', {1,oC}, {0.5, 0.5}, typeid(TypeParam) == typeid(float) ? nd4j::DataType::FLOAT32 : nd4j::DataType::DOUBLE);

    NDArray gradI(&input);
    NDArray gradWD(&weightsDepth);
    NDArray gradWP(&weightsPoint);
    NDArray gradB(&bias);

    input = 2.;
    weightsDepth.linspace(0.1, 0.1);
    weightsPoint.linspace(0.15, 0.1);
    gradO.linspace(0.01, 0.01);

    nd4j::ops::sconv2d_bp op;
    Nd4jStatus status = op.execute({&input, &gradO, &weightsDepth, & weightsPoint, &bias},
                              {&gradI, &gradWD, &gradWP, &gradB},
                              {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat}, {});

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

    NDArray expGradI  = gradI;
    NDArray expGradWD = gradWD;
    NDArray expGradWP = gradWP;
    NDArray expGradB  = gradB;

    for( int i=0; i<10; i++ ) {
        Nd4jStatus status = op.execute({&input, &gradO, &weightsDepth, & weightsPoint, &bias},
                              {&gradI, &gradWD, &gradWP, &gradB},
                              {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat}, {});
        ASSERT_EQ(Status::OK(), status);

        ASSERT_TRUE(expGradI.equalsTo(gradI));
        ASSERT_TRUE(expGradWD.equalsTo(gradWD));
        ASSERT_TRUE(expGradWP.equalsTo(gradWP));
        ASSERT_TRUE(expGradB.equalsTo(expGradB));
    }
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, sconv2d_bp_3) {

    auto input = NDArrayFactory::create<TypeParam>('c', {3, 3, 16, 16});
    auto weightsD = NDArrayFactory::create<TypeParam>('c', {1, 3, 2, 2});
    auto weightsP = NDArrayFactory::create<TypeParam>('c', {2, 3, 1, 1});
    auto bias = NDArrayFactory::create<TypeParam>('c', {1, 2});

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

    auto epsilonNext = NDArrayFactory::create<TypeParam>('c', {3, 2, 14, 14});

    auto epsilon = NDArrayFactory::create<TypeParam>('c', {3, 3, 16, 16});

    nd4j::ops::sconv2d_bp op;
    auto result = op.execute({&input, &epsilonNext, &weightsD, &weightsP}, {}, {2, 2, 1, 1, 0, 0, 2, 2, 0});

    auto eps = result->at(0);
    auto gWD = result->at(1);
    auto gWP = result->at(2);


    ASSERT_TRUE(epsilon.isSameShape(eps));

    delete result;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, sconv2d_bp_4) {

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

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

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

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

    nd4j::ops::sconv2d_bp op;
    auto results = op.execute({&input, &gradO, &weightsDepth, &bias}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat});
    auto* gradI = results->at(0);
    auto* gradWD = results->at(1);

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

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

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, sconv2d_bp_5) {

    int bS=1, iH=8,iW=8,  iC=3,mC=3,  kH=1,kW=1,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=8,oW=8;
    int       oC=2;                  // iC*mC if weightsPoint = nullptr
    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 gradO    = NDArrayFactory::create<double>('c', {bS, oC, oH, oW});
    auto weightsDepth  = NDArrayFactory::create<double>('c', {kH, kW, iC, mC});
    auto weightsPoint  = NDArrayFactory::create<double>('c', {1, 1, iC*mC, oC});
    auto bias     = NDArrayFactory::create<double>('c', {1,oC}, {1,2});

    auto gradI   = NDArrayFactory::create<double>('c', {bS, iC, iH, iW});
    auto gradWD  = NDArrayFactory::create<double>('f', {kH, kW, iC, mC});
    auto gradWP  = NDArrayFactory::create<double>('c', {1, 1, iC*mC, oC});
    auto gradB   = NDArrayFactory::create<double>('c', {1,oC}, {1,2});

    input = 2.;
    weightsDepth.linspace(0.1, 0.1);
    weightsDepth.linspace(-0.5, 0.1);
    gradO.linspace(0.01, 0.01);

    nd4j::ops::sconv2d_bp op;
    auto status = op.execute({&input, &gradO, &weightsDepth, &weightsPoint, &bias}, {&gradI, &gradWD, &gradWP, &gradB}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, paddingMode, dataFormat}, {});
    ASSERT_EQ(Status::OK(), status);
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, im2col_bp_1) {

    int bS=3, iH=12,iW=12,  iC=6,oC=3,  kH=2,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=12,oW=12;

    // [bS, iC, kH, kW, oH, oW] is de-convoluted to [bS, iC, iH, iW]
    NDArray input('c', {bS, iC, iH, iW}, nd4j::DataType::DOUBLE);
    NDArray gradO('c', {bS, iC, kH, kW, oH, oW}, nd4j::DataType::DOUBLE);
    NDArray gradI('c', {bS, iC, iH, iW}, nd4j::DataType::DOUBLE);           // output

    nd4j::ops::im2col_bp op;
    Nd4jStatus status = op.execute({&input, &gradO}, {&gradI}, {}, {kH, kW, sH, sW, pH, pW, dH, dW, 1}, {});

    ASSERT_EQ(ND4J_STATUS_OK, status);

}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, deconv3d_test1) {

    int bS=2, iD=4,iH=4,iW=4,  iC=2,oC=3,  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=3,oW=3;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<double>('c', {bS, oD, oH, oW, oC});
    auto weights  = NDArrayFactory::create<double>('c', {kD, kH, kW, iC, oC});
    auto exp = NDArrayFactory::create<double>('c', {bS, iD, iH, iW, iC}, {0.3 , 0.75, 1.5 , 2.4 , 1.5 , 2.4 , 1.2 , 1.65, 2.4 , 3.3 , 6.6 , 8.4 , 6.6 , 8.4 , 4.2 , 5.1 , 2.4 , 3.3 , 6.6 , 8.4 , 6.6 , 8.4 , 4.2 , 5.1 , 2.1 , 2.55, 5.1 , 6.  , 5.1 , 6.  , 3.  , 3.45,
                                                    4.2 , 5.1 ,10.2 ,12.  ,10.2 ,12.  , 6.  , 6.9 ,12.  ,13.8 ,27.6 ,31.2 ,27.6 ,31.2 ,15.6 ,17.4 ,12.  ,13.8 ,27.6 ,31.2 ,27.6 ,31.2 ,15.6 ,17.4 , 7.8 , 8.7 ,17.4 ,19.2 ,17.4 ,19.2 , 9.6 ,10.5 ,
                                                    4.2 , 5.1 ,10.2 ,12.  ,10.2 ,12.  , 6.  , 6.9 ,12.  ,13.8 ,27.6 ,31.2 ,27.6 ,31.2 ,15.6 ,17.4 ,12.  ,13.8 ,27.6 ,31.2 ,27.6 ,31.2 ,15.6 ,17.4 , 7.8 , 8.7 ,17.4 ,19.2 ,17.4 ,19.2 , 9.6 ,10.5 ,
                                                    3.9 , 4.35, 8.7 , 9.6 , 8.7 , 9.6 , 4.8 , 5.25, 9.6 ,10.5 ,21.  ,22.8 ,21.  ,22.8 ,11.4 ,12.3 , 9.6 ,10.5 ,21.  ,22.8 ,21.  ,22.8 ,11.4 ,12.3 , 5.7 , 6.15,12.3 ,13.2 ,12.3 ,13.2 , 6.6 , 7.05,
                                                    0.3 , 0.75, 1.5 , 2.4 , 1.5 , 2.4 , 1.2 , 1.65, 2.4 , 3.3 , 6.6 , 8.4 , 6.6 , 8.4 , 4.2 , 5.1 , 2.4 , 3.3 , 6.6 , 8.4 , 6.6 , 8.4 , 4.2 , 5.1 , 2.1 , 2.55, 5.1 , 6.  , 5.1 , 6.  , 3.  , 3.45,
                                                    4.2 , 5.1 ,10.2 ,12.  ,10.2 ,12.  , 6.  , 6.9 ,12.  ,13.8 ,27.6 ,31.2 ,27.6 ,31.2 ,15.6 ,17.4 ,12.  ,13.8 ,27.6 ,31.2 ,27.6 ,31.2 ,15.6 ,17.4 , 7.8 , 8.7 ,17.4 ,19.2 ,17.4 ,19.2 , 9.6 ,10.5 ,
                                                    4.2 , 5.1 ,10.2 ,12.  ,10.2 ,12.  , 6.  , 6.9 ,12.  ,13.8 ,27.6 ,31.2 ,27.6 ,31.2 ,15.6 ,17.4 ,12.  ,13.8 ,27.6 ,31.2 ,27.6 ,31.2 ,15.6 ,17.4 , 7.8 , 8.7 ,17.4 ,19.2 ,17.4 ,19.2 , 9.6 ,10.5 ,
                                                    3.9 , 4.35, 8.7 , 9.6 , 8.7 , 9.6 , 4.8 , 5.25, 9.6 ,10.5 ,21.  ,22.8 ,21.  ,22.8 ,11.4 ,12.3 , 9.6 ,10.5 ,21.  ,22.8 ,21.  ,22.8 ,11.4 ,12.3 , 5.7 , 6.15,12.3 ,13.2 ,12.3 ,13.2 , 6.6 , 7.05});
    input = 0.5;
    weights.linspace(0.1, 0.1);

    nd4j::ops::deconv3d 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);

    // output->printBuffer();

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, deconv3d_test2) {

    int bS=2, iD=4,iH=4,iW=4,  iC=2,oC=3,  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=4,oH=4,oW=4;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<double>('c', {bS, oD, oH, oW, oC});
    auto weights  = NDArrayFactory::create<double>('c', {kD, kH, kW, iC, oC});
    auto exp = NDArrayFactory::create<double>('c', {bS, iD, iH, iW, iC}, {0.3 ,  0.75, 1.5 ,  2.4 , 1.5 ,  2.4 , 1.5 ,  2.4 , 2.4 ,  3.3 , 6.6 ,  8.4 , 6.6 ,  8.4 , 6.6 ,  8.4 , 2.4 ,  3.3 , 6.6 ,  8.4 , 6.6 ,  8.4 , 6.6 ,  8.4 , 2.4 ,  3.3 , 6.6 ,  8.4 , 6.6 ,  8.4 , 6.6 ,  8.4 ,
                                                    4.2 ,  5.1 ,10.2 , 12.  ,10.2 , 12.  ,10.2 , 12.  ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,
                                                    4.2 ,  5.1 ,10.2 , 12.  ,10.2 , 12.  ,10.2 , 12.  ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,
                                                    4.2 ,  5.1 ,10.2 , 12.  ,10.2 , 12.  ,10.2 , 12.  ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,
                                                    0.3 ,  0.75, 1.5 ,  2.4 , 1.5 ,  2.4 , 1.5 ,  2.4 , 2.4 ,  3.3 , 6.6 ,  8.4 , 6.6 ,  8.4 , 6.6 ,  8.4 , 2.4 ,  3.3 , 6.6 ,  8.4 , 6.6 ,  8.4 , 6.6 ,  8.4 , 2.4 ,  3.3 , 6.6 ,  8.4 , 6.6 ,  8.4 , 6.6 ,  8.4 ,
                                                    4.2 ,  5.1 ,10.2 , 12.  ,10.2 , 12.  ,10.2 , 12.  ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,
                                                    4.2 ,  5.1 ,10.2 , 12.  ,10.2 , 12.  ,10.2 , 12.  ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,
                                                    4.2 ,  5.1 ,10.2 , 12.  ,10.2 , 12.  ,10.2 , 12.  ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 ,12.  , 13.8 ,27.6 , 31.2 ,27.6 , 31.2 ,27.6 , 31.2 });
    input = 0.5;
    weights.linspace(0.1, 0.1);

    nd4j::ops::deconv3d 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(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, deconv3d_test3) {

    int bS=2, iD=4,iH=4,iW=4,  iC=2,oC=3,  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=3,oW=3;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<double>('c', {bS, oC, oD, oH, oW});
    auto weights  = NDArrayFactory::create<double>('c', {oC, iC, kD, kH, kW});
    auto exp = NDArrayFactory::create<double>('c', {bS, iC, iD, iH, iW}, {2.55,  5.25,  5.25,  2.7, 5.4 , 11.1 , 11.1 ,  5.7, 5.4 , 11.1 , 11.1 ,  5.7, 2.85,  5.85,  5.85,  3. , 5.7 , 11.7 , 11.7 ,  6. ,12.  , 24.6 , 24.6 , 12.6,12.  , 24.6 , 24.6 , 12.6, 6.3 , 12.9 , 12.9 ,  6.6,
                                                    5.7 , 11.7 , 11.7 ,  6. ,12.  , 24.6 , 24.6 , 12.6,12.  , 24.6 , 24.6 , 12.6, 6.3 , 12.9 , 12.9 ,  6.6, 3.15,  6.45,  6.45,  3.3, 6.6 , 13.5 , 13.5 ,  6.9, 6.6 , 13.5 , 13.5 ,  6.9, 3.45,  7.05,  7.05,  3.6,
                                                    3.75,  7.65,  7.65,  3.9, 7.8 , 15.9 , 15.9 ,  8.1, 7.8 , 15.9 , 15.9 ,  8.1, 4.05,  8.25,  8.25,  4.2, 8.1 , 16.5 , 16.5 ,  8.4,16.8 , 34.2 , 34.2 , 17.4,16.8 , 34.2 , 34.2 , 17.4, 8.7 , 17.7 , 17.7 ,  9. ,
                                                    8.1 , 16.5 , 16.5 ,  8.4,16.8 , 34.2 , 34.2 , 17.4,16.8 , 34.2 , 34.2 , 17.4, 8.7 , 17.7 , 17.7 ,  9. , 4.35,  8.85,  8.85,  4.5, 9.  , 18.3 , 18.3 ,  9.3, 9.  , 18.3 , 18.3 ,  9.3, 4.65,  9.45,  9.45,  4.8,
                                                    2.55,  5.25,  5.25,  2.7, 5.4 , 11.1 , 11.1 ,  5.7, 5.4 , 11.1 , 11.1 ,  5.7, 2.85,  5.85,  5.85,  3. , 5.7 , 11.7 , 11.7 ,  6. ,12.  , 24.6 , 24.6 , 12.6,12.  , 24.6 , 24.6 , 12.6, 6.3 , 12.9 , 12.9 ,  6.6,
                                                    5.7 , 11.7 , 11.7 ,  6. ,12.  , 24.6 , 24.6 , 12.6,12.  , 24.6 , 24.6 , 12.6, 6.3 , 12.9 , 12.9 ,  6.6, 3.15,  6.45,  6.45,  3.3, 6.6 , 13.5 , 13.5 ,  6.9, 6.6 , 13.5 , 13.5 ,  6.9, 3.45,  7.05,  7.05,  3.6,
                                                    3.75,  7.65,  7.65,  3.9, 7.8 , 15.9 , 15.9 ,  8.1, 7.8 , 15.9 , 15.9 ,  8.1, 4.05,  8.25,  8.25,  4.2, 8.1 , 16.5 , 16.5 ,  8.4,16.8 , 34.2 , 34.2 , 17.4,16.8 , 34.2 , 34.2 , 17.4, 8.7 , 17.7 , 17.7 ,  9. ,
                                                    8.1 , 16.5 , 16.5 ,  8.4,16.8 , 34.2 , 34.2 , 17.4,16.8 , 34.2 , 34.2 , 17.4, 8.7 , 17.7 , 17.7 ,  9. , 4.35,  8.85,  8.85,  4.5, 9.  , 18.3 , 18.3 ,  9.3, 9.  , 18.3 , 18.3 ,  9.3, 4.65,  9.45,  9.45,  4.8});
    input = 0.5;
    weights.linspace(0.1, 0.1);
    weights.permutei({2, 3, 4, 1, 0});

    nd4j::ops::deconv3d 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(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, deconv3d_test4) {

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

    auto input    = NDArrayFactory::create<double>('c', {bS, oC, oD, oH, oW});
    auto weights  = NDArrayFactory::create<double>('c', {oC, iC, kD, kH, kW});
    auto exp = NDArrayFactory::create<double>('c', {bS, iC, iD, iH, iW}, {24.6, 24.6,24.6, 24.6,24.6, 24.6,24.6, 24.6,34.2, 34.2,34.2, 34.2,34.2, 34.2,34.2, 34.2,24.6, 24.6,24.6, 24.6,
                                                    24.6, 24.6,24.6, 24.6,34.2, 34.2,34.2, 34.2,34.2, 34.2,34.2, 34.2});
    input = 0.5;
    weights.linspace(0.1, 0.1);
    weights.permutei({2, 3, 4, 1, 0});

    nd4j::ops::deconv3d 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(exp.isSameShape(output));
    ASSERT_TRUE(exp.equalsTo(output));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, deconv3d_test5) {
    int bS=1, oD=5,oH=5,oW=5,  oC=3,iC=2,  kD=2,kH=2,kW=2,  sD=1,sH=1,sW=1,  pD=0,pH=0,pW=0,  dD=2,dH=2,dW=2;
    int       iD=3,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, iD, iH, iW, iC});
    auto weights  = NDArrayFactory::create<float>('c', {kD, kH, kW, oC, iC});
    auto bias     = NDArrayFactory::create<float>('c', {oC});

    auto exp = NDArrayFactory::create<float>('c', {bS, oD, 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, -0.2f, -0.5f, -0.8f, 0.1f, 0.2f, 0.3f, -0.7f, -0.5f, -0.3f, 0.4f, 0.5f, 0.6f, 1.9f, 2.4f,
                2.9f, 0.7f, 1.6f, 2.5f, 1.0f, 2.3f, 3.6f, 4.7f, 7.3f, 9.9f, 4.9f, 6.2f, 7.5f, 6.4f, 8.1f, 9.8f, -0.4f, 1.4f, 3.2f, 2.6f, 5.2f, 7.8f, 10.6f, 15.8f, 21.0f, 10.4f, 13.0f, 15.6f,
                15.8f, 19.2f, 22.6f, 6.1f, 7.0f, 7.9f, 8.8f, 10.1f, 11.4f, 20.3f, 22.9f, 25.5f, 12.7f, 14.0f, 15.3f, 16.6f, 18.3f, 20.0f, 14.2f, 16.3f, 18.4f, 16.9f, 19.4f, 21.9f, 40.1f,
                45.1f, 50.1f, 24.4f, 26.9f, 29.4f, 28.3f, 31.2f, 34.1f, -47.2f, -47.8f, -48.4f, -41.8f, -41.6f, -41.4f, -85.4f, -85.f, -84.6f, -41.2f, -41.0f, -40.8f, -33.4f, -32.4f, -31.4f,
                -31.f, -29.2f, -27.4f, -25.6f, -23.0f, -20.4f, -45.8f, -40.6f, -35.4f, -17.8f, -15.2f, -12.6f, -10.0f, -6.6f, -3.2f, -65.6f, -62.0f, -58.4f, -50.0f, -44.8f, -39.6f, -89.2f,
                -78.8f, -68.4f, -34.4f, -29.2f, -24.f, -14.0f, -7.2f, -0.4f, -20.2f, -18.4f, -16.6f, -10.f, -7.4f, -4.8f, -14.6f, -9.4f, -4.2f, -2.2f, 0.4f, 3.0f, 10.4f, 13.8f, 17.2f, 10.4f,
                14.6f, 18.8f, 20.6f, 25.6f, 30.6f, 53.8f, 63.8f, 73.8f, 35.6f, 40.6f, 45.6f, 48.2f, 54.0f, 59.8f, -3.8f, -4.1f, -4.4f, 1.3f, 1.4f, 1.5f, 1.7f, 1.9f, 2.1f, 1.6f, 1.7f, 1.8f, 7.9f,
                8.4f, 8.9f, 11.5f, 12.4f, 13.3f, 16.6f, 17.9f, 19.2f, 35.9f, 38.5f, 41.1f, 20.5f, 21.8f, 23.1f, 26.8f, 28.5f, 30.2f, 21.2f, 23.0f, 24.8f, 33.8f, 36.4f, 39.0f, 73.0f, 78.2f,
                83.4f, 41.6f, 44.2f, 46.8f, 56.6f, 60.0f, 63.4f, 16.9f, 17.8f, 18.7f, 24.4f, 25.7f, 27.f, 51.5f, 54.1f, 56.7f, 28.3f, 29.6f, 30.9f, 37.0f, 38.7f, 40.4f, 39.4f, 41.5f,
                43.6f, 46.9f, 49.4f, 51.9f, 100.1f, 105.1f, 110.1f, 54.4f, 56.9f, 59.4f, 63.1f, 66.0f, 68.9f, 42.1f, 45.4f, 48.7f, 47.2f, 50.9f, 54.6f, 104.3f, 111.7f,
                119.1f, 58.3f, 62.0f, 65.7f, 64.6f, 68.7f, 72.8f, 57.4f, 61.9f, 66.4f, 62.5f, 67.4f, 72.3f, 138.5f, 148.3f, 158.1f, 77.2f, 82.1f, 87.0f, 83.5f, 88.8f, 94.1f,
                134.6f, 143.6f, 152.6f, 147.2f, 157.0f, 166.8f, 321.4f, 341.0f, 360.6f, 176.6f, 186.4f, 196.2f, 191.6f, 202.2f, 212.8f, 84.4f, 88.9f,
                93.4f, 91.9f, 96.8f, 101.7f, 197.3f, 207.1f, 216.9f, 106.6f, 111.5f, 116.4f, 115.3f, 120.6f, 125.9f, 106.9f, 112.6f, 118.3f, 114.4f, 120.5f, 126.6f, 245.9f, 258.1f, 270.3f, 132.7f, 138.8f, 144.9f, 141.4f, 147.9f, 154.4f});

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

    nd4j::ops::deconv3d op;
    auto results = op.execute({&input, &weights}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,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(ConvolutionTests2, deconv3d_bp_test1) {

    int bS=1, iD=3,iH=3,iW=3,  iC=1,oC=2,  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=2,oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

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

    NDArray expGradI('c', {bS, oD, oH, oW, oC}, {62., 67.6, 68.4, 74.8, 81.2, 89.2, 87.6, 96.4, 119.6, 132.4, 126., 139.6, 138.8, 154., 145.2, 161.2}, nd4j::DataType::FLOAT32);
    NDArray expGradW('c', {kD, kH, kW, iC, oC}, {28., 28., 32., 32., 40., 40., 44., 44., 64, 64., 68., 68., 76., 76., 80., 80.}, nd4j::DataType::FLOAT32);
    NDArray expGradB('c', {iC}, {364.5}, nd4j::DataType::FLOAT32);

    input = 0.5;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.5);

    nd4j::ops::deconv3d_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;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, deconv3d_bp_test2) {

    int bS=1, iD=2,iH=2,iW=2,  iC=1,oC=2,  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=2,oH=2,oW=2;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

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

    NDArray expGradI('c', {bS, oD, oH, oW, oC}, {34, 37.2, 16.6, 18.4, 15.4, 17.4, 7.1, 8.2, 10.6, 13., 4.3, 5.6, 2.9, 4.3, 0.75, 1.5}, nd4j::DataType::FLOAT32);
    NDArray expGradW('c', {kD, kH, kW, iC, oC}, {16, 16, 9, 9, 10, 10, 5.5, 5.5, 12, 12, 6.5, 6.5, 7, 7, 3.75, 3.75}, nd4j::DataType::FLOAT32);

    input = 0.5;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.5);

    nd4j::ops::deconv3d_bp op;
    auto results = op.execute({&input, &weights, &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;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, deconv3d_bp_test3) {

    int bS=1, iD=3,iH=3,iW=3,  iC=1,oC=2,  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=2,oH=2,oW=2;
    int paddingMode = 0;             // 1-SAME, 0-VALID;
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<float>('c', {bS, oC, oD, oH, oW});
    auto weights  = NDArrayFactory::create<float>('c', {kD, kH, kW, iC, oC}, {0.1f, 0.9f, 0.2f, 0.1f, 0.3f, 1.1f, 0.4f, 1.2f, 0.5f, 1.3f, 0.6f, 1.4f, 0.7f, 1.5f, 0.8f, 1.6f});
    auto gradO    = NDArrayFactory::create<float>('c', {bS, iC, iD, iH, iW});

    NDArray expGradI('c', {bS, oD, oH, oW, oC}, {33.8, 37.4, 44.6, 48.2, 66.2, 69.8, 77., 80.6, 77.25, 86.35, 104.55, 113.65, 159.15, 168.25, 186.45, 195.55}, nd4j::DataType::FLOAT32);
    NDArray expGradW('c', {kD, kH, kW, iC, oC}, {28., 28, 32, 32, 40, 40, 44, 44, 64, 64, 68, 68, 76, 76, 80, 80.}, nd4j::DataType::FLOAT32);

    input = 0.5;
    gradO.linspace(0.5);

    nd4j::ops::deconv3d_bp op;
    auto results = op.execute({&input, &weights, &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;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, deconv3d_bp_test4) {

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

    auto input    = NDArrayFactory::create<float>('c', {bS, oC, oD, oH, oW});
    auto weights  = NDArrayFactory::create<float>('c', {kD, kH, kW, iC, oC}, {0.1f, 0.9f, 0.2f, 0.1f, 0.3f, 1.1f, 0.4f, 1.2f, 0.5f, 1.3f, 0.6f, 1.4f, 0.7f, 1.5f, 0.8f, 1.6f});
    auto gradO    = NDArrayFactory::create<float>('c', {bS, iC, iD, iH, iW});

    NDArray expGradI('c', {bS, oC, oD, oH, oW}, {0.4, 1.55, 1.05, 2.3, 5.7, 3.2, 1.5, 3.35, 1.75, 3.8, 8.3, 4.3, 9.0, 18.6, 9.2, 4.4, 8.7, 4.1, 1.8, 3.55, 1.65, 3.5, 6.5, 2.8, 1.3, 2.15, 0.75, 0.8, 3.15, 2.25, 4.7, 12.1, 7.2, 3.5, 8.15, 4.55, 7.8, 17.9, 9.9, 19.75, 42.85, 23.6, 9.35, 21.55, 12.9, 5.4, 11.55, 6.05, 8.25, 20.75, 13.2, 0.65, 6.6, 6.75}, nd4j::DataType::FLOAT32);
    NDArray expGradW('c', {kD, kH, kW, iC, oC}, {16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.0, 16.}, nd4j::DataType::FLOAT32);

    input = 0.5;
    gradO.linspace(0.5);

    nd4j::ops::deconv3d_bp op;
    auto results = op.execute({&input, &weights, &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;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, maxpool2d_1) {

    auto x = NDArrayFactory::create_<float>('c', {bS,iD,iH,iW});
    auto exp = NDArrayFactory::create<float>('c',{bS,iD,oH,oW});
    // auto z('c',{bS,iD,oH,oW});

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, x);
    // variableSpace->putVariable(1, &z);

    auto block = new Context(1, variableSpace, false);
    block->fillInputs({-1});
    std::vector<int>* argI = block->getIArguments();
    *argI = {kH,kW, sH,sW, pH,pW, dH,dW, 0};  // 0,1 - kernel Height/Width; 2,3 - stride Height/Width; 4,5 - pad Height/Width; 6,7 - dilation Height/Width; 8 - same mode;

    nd4j::ops::maxpool2d pooling;
    Nd4jStatus status = pooling.execute(block);
    ASSERT_EQ(ND4J_STATUS_OK, status);

    auto result = variableSpace->getVariable(block->getNodeId())->getNDArray();
    // result->printShapeInfo();
    ASSERT_TRUE(exp.isSameShape(result));

    delete variableSpace;
    delete block;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, maxpool2d_2) {

    const int bS = 2;
    const int iD = 1;
    const int iH = 28;
    const int iW = 28;
    const int kH = 5;
    const int kW = 5;
    const int sH = 1;
    const int sW = 1;
    const int pH = 0;
    const int pW = 0;
    const int dH = 1;
    const int dW = 1;
    const int oH = (iH - kH - (kH-1)*(dH-1) + 2*pH)/sH + 1;     // output height
    const int oW = (iW - kW - (kW-1)*(dW-1) + 2*pW)/sW + 1;     // output width


    auto x = NDArrayFactory::create_<float>('c', {bS,iD,iH,iW});
    auto exp = NDArrayFactory::create<float>('c',{bS,iD,oH,oW});
    // auto z('c',{bS,iD,oH,oW});

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, x);
    // variableSpace->putVariable(1, &z);

    auto block = new Context(1, variableSpace, false);
    block->fillInputs({-1});
    std::vector<int>* argI = block->getIArguments();
    *argI = {kH,kW, sH,sW, pH,pW, dH,dW, 0};  // 0,1 - kernel Height/Width; 2,3 - stride Height/Width; 4,5 - pad Height/Width; 6,7 - dilation Height/Width; 8 - same mode;

    nd4j::ops::maxpool2d pooling;
    Nd4jStatus status = pooling.execute(block);
    ASSERT_EQ(ND4J_STATUS_OK, status);

    auto result = variableSpace->getVariable(block->getNodeId())->getNDArray();
    // result->printShapeInfo();
    ASSERT_TRUE(exp.isSameShape(result));

    delete variableSpace;
    delete block;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, maxpool2d_3) {

    const int bS = 2;
    const int iD = 1;
    const int iH = 28;
    const int iW = 28;
    const int kH = 5;
    const int kW = 5;
    const int sH = 1;
    const int sW = 1;
    const int pH = 0;
    const int pW = 0;
    const int dH = 1;
    const int dW = 1;
    const int oH = (int) nd4j::math::nd4j_ceil<float, int>(iH * 1.f / sH);
    const int oW = (int) nd4j::math::nd4j_ceil<float, int>(iW * 1.f / sW);


    auto x = NDArrayFactory::create_<float>('c', {bS,iD,iH,iW});
    auto exp = NDArrayFactory::create<float>('c',{bS,iD,oH,oW});
    // auto z('c',{bS,iD,oH,oW});

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, x);
    // variableSpace->putVariable(1, &z);

    auto block = new Context(1, variableSpace, false);
    block->fillInputs({-1});
    std::vector<int>* argI = block->getIArguments();
    *argI = {kH,kW, sH,sW, pH,pW, dH,dW, 1};  // 0,1 - kernel Height/Width; 2,3 - stride Height/Width; 4,5 - pad Height/Width; 6,7 - dilation Height/Width; 8 - same mode;

    nd4j::ops::maxpool2d pooling;
    Nd4jStatus status = pooling.execute(block);
    ASSERT_EQ(ND4J_STATUS_OK, status);

    auto result = variableSpace->getVariable(block->getNodeId())->getNDArray();
    // result->printShapeInfo();
    ASSERT_TRUE(exp.isSameShape(result));

    delete variableSpace;
    delete block;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, maxpool2d_4) {

    const int bS = 2;
    const int iD = 1;
    const int iH = 24;
    const int iW = 24;
    const int kH = 3;
    const int kW = 3;
    const int sH = 1;
    const int sW = 1;
    const int pH = 0;
    const int pW = 0;
    const int dH = 1;
    const int dW = 1;
    const int oH = (iH - kH - (kH-1)*(dH-1) + 2*pH)/sH + 1;     // output height
    const int oW = (iW - kW - (kW-1)*(dW-1) + 2*pW)/sW + 1;     // output width


    auto x = NDArrayFactory::create_<float>('c', {bS,iD,iH,iW});
    auto exp = NDArrayFactory::create<float>('c',{bS,iD,oH,oW});
    // auto z('c',{bS,iD,oH,oW});

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, x);
    // variableSpace->putVariable(1, &z);

    auto block = new Context(1, variableSpace, false);
    block->fillInputs({-1});
    std::vector<int>* argI = block->getIArguments();
    *argI = {kH,kW, sH,sW, pH,pW, dH,dW, 0};  // 0,1 - kernel Height/Width; 2,3 - stride Height/Width; 4,5 - pad Height/Width; 6,7 - dilation Height/Width; 8 - same mode;

    nd4j::ops::maxpool2d pooling;
    Nd4jStatus status = pooling.execute(block);
    ASSERT_EQ(ND4J_STATUS_OK, status);

    auto result = variableSpace->getVariable(block->getNodeId())->getNDArray();
    // result->printShapeInfo();
    ASSERT_TRUE(exp.isSameShape(result));

    delete variableSpace;
    delete block;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, maxpool2d_5) {

    const int bS = 2;
    const int iD = 1;
    const int iH = 24;
    const int iW = 24;
    const int kH = 3;
    const int kW = 3;
    const int sH = 1;
    const int sW = 1;
    const int pH = 0;
    const int pW = 0;
    const int dH = 1;
    const int dW = 1;
    const int oH = (int) nd4j::math::nd4j_ceil<float, int>(iH * 1.f / sH);
    const int oW = (int) nd4j::math::nd4j_ceil<float, int>(iW * 1.f / sW);


    auto x = NDArrayFactory::create_<float>('c', {bS,iD,iH,iW});
    auto exp = NDArrayFactory::create<float>('c',{bS,iD,oH,oW});
    // auto z('c',{bS,iD,oH,oW});

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, x);
    // variableSpace->putVariable(1, &z);

    auto block = new Context(1, variableSpace, false);
    block->fillInputs({-1});
    std::vector<int>* argI = block->getIArguments();
    *argI = {kH,kW, sH,sW, pH,pW, dH,dW, 1};  // 0,1 - kernel Height/Width; 2,3 - stride Height/Width; 4,5 - pad Height/Width; 6,7 - dilation Height/Width; 8 - same mode;

    nd4j::ops::maxpool2d pooling;
    Nd4jStatus status = pooling.execute(block);
    ASSERT_EQ(ND4J_STATUS_OK, status);

    auto result = variableSpace->getVariable(block->getNodeId())->getNDArray();
    // result->printShapeInfo();
    ASSERT_TRUE(exp.isSameShape(result));

    delete variableSpace;
    delete block;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, maxpool2d_6) {
    auto x = NDArrayFactory::create<TypeParam>('c', {2, 4, 4, 2});
    auto exp = NDArrayFactory::create<TypeParam>('c', {2, 2, 2, 2}, {11.f,  12.f,  15.f,  16.f,  27.f,  28.f,  31.f,  32.f,  43.f,  44.f,  47.f,  48.f,  59.f,  60.f,  63.f, 64.f});

    x.linspace(1);

    nd4j::ops::maxpool2d op;
    auto result = op.execute({&x}, {}, {2, 2, 2, 2, 0, 0, 1, 1, 1, 1, 1});

    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(TypedConvolutionTests2, maxpool2d_7) {
    auto x = NDArrayFactory::create<TypeParam>('c', {2, 4, 4, 2});
    auto exp = NDArrayFactory::create<TypeParam>('c', {2, 2, 2, 2}, {11.f,  12.f,  15.f,  16.f,  27.f,  28.f,  31.f,  32.f,  43.f,  44.f,  47.f,  48.f,  59.f,  60.f,  63.f, 64.f});

    x.linspace(1);

    nd4j::ops::maxpool2d op;
    auto result = op.execute({&x}, {}, {2, 2, 2, 2, 0, 0, 1, 1, 0, 1, 1});

    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(TypedConvolutionTests2, maxpool2d_8) {
    auto x = NDArrayFactory::create<TypeParam>('c', {2, 2, 5, 5});
    auto exp = NDArrayFactory::create<TypeParam>('c', {2, 2, 2, 2}, {7.f, 9.f, 17.f, 19.f, 32.f, 34.f, 42.f, 44.f, 57.f, 59.f, 67.f, 69.f, 82.f, 84.f, 92.f, 94.f});

    x.linspace(1);

    nd4j::ops::maxpool2d op;
    auto result = op.execute({&x}, {}, {2, 2, 2, 2, 0, 0, 1, 1, 0, 1, 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(TypedConvolutionTests2, maxpool2d_9) {

    int bS = 3;                 // batch size (number of samples)
    int iC = 3;                 // input channels
    int iH = 28, iW = 28;       // input height/width
    int kH = 2,  kW = 2;        // kernel (filter) height/width
    int sH = 1,  sW = 1;        // stride height/width
    int pH = 0,  pW = 0;        // padding height/width
    int dH = 1,  dW = 1;        // dilation height/width

    int oH = 27, oW = 27;       // output height/width

    int isSameMode = 0;         // 1-SAME,  0-VALID

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

    nd4j::ops::maxpool2d op;
    auto results = op.execute({&input}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW, isSameMode, 1, 0});
    auto output = results->at(0);

    ASSERT_EQ(Status::OK(), results->status());
    ASSERT_TRUE(output->isSameShape({bS, iC, oH, oW}));

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, maxpool2d_10) {

    int bS=1, iH=4,iW=4,  iC=3,  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  = 0;             // 1-NHWC, 0-NCHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW}, {0.27620894f, 0.21801452f, 0.062078513f, 7.348895E-4f, 0.24149609f, 0.4948205f, 0.93483436f, 0.52035654f, 0.30292067f, 0.3289706f, 0.7977864f,
                                                     0.03180518f, 0.1455722f, 0.90352905f, 0.9405744f, 0.0048329555f, 0.44062102f, 0.111197524f, 0.31742015f, 0.1933705f, 0.23825112f, 0.35076278f, 0.7135856f, 0.28229436f, 0.18310733f,
                                                     0.9613717f, 0.56823575f, 0.78289545f, 0.62195826f, 0.5244586f, 0.5040889f, 0.025349546f, 0.41400263f, 0.28420195f, 0.8536445f, 0.3044107f, 0.7997134f, 0.45762005f, 0.7653578f,
                                                     0.07198584f, 0.5304998f, 0.7334402f, 0.85019743f, 0.031957153f, 0.37088063f, 0.85722464f, 0.06376881f, 0.39791203f});

    auto expOutput = NDArrayFactory::create<TypeParam>('c', {bS, iC, oH, oW}, {0.4948205f, 0.93483436f, 0.93483436f, 0.4948205f, 0.93483436f, 0.93483436f, 0.90352905f, 0.9405744f, 0.9405744f, 0.44062102f, 0.7135856f,
                                                     0.7135856f, 0.9613717f, 0.9613717f, 0.78289545f, 0.9613717f, 0.9613717f, 0.78289545f, 0.7997134f, 0.8536445f, 0.8536445f, 0.7997134f, 0.85019743f, 0.85019743f,
                                                     0.85722464f, 0.85722464f, 0.85019743f});

    nd4j::ops::maxpool2d op;
    auto results = op.execute({&input}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode});
    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(ConvolutionTests2, maxpool2d_11) {

    NDArray input('c', {1,1,4,5}, nd4j::DataType::FLOAT32);
    NDArray z('c', {1,1,4,5}, nd4j::DataType::FLOAT32);

    input.linspace(1.);

    nd4j::ops::maxpool2d op;
    auto results = op.execute({&input}, {}, {2,2,  1,1,  1,1,  2,2,  1,0,0});

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, avgpool3d_test1) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, oD, oH, oW}, {10.5f, 11.5f, 13.5f, 14.5f, 22.5f, 23.5f, 25.5f, 26.5f, 46.5f, 47.5f, 49.5f, 50.5f, 58.5f, 59.5f, 61.5f, 62.5f,
                                                         82.5f, 83.5f, 85.5f, 86.5f, 94.5f, 95.5f, 97.5f, 98.5f,118.5f,119.5f,121.5f,122.5f,130.5f,131.5f,133.5f,134.5f,
                                                        154.5f,155.5f,157.5f,158.5f,166.5f,167.5f,169.5f,170.5f,190.5f,191.5f,193.5f,194.5f,202.5f,203.5f,205.5f,206.5f});
    input.linspace(1.);

    nd4j::ops::avgpool3dnew op;
    auto results = op.execute({&input}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW, dD,dH,dW, paddingMode, 1, 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(TypedConvolutionTests2, avgpool3d_test2) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 expected = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, iC}, {  25.f, 26.f, 27.f, 28.f, 29.f, 30.f, 29.5f, 30.5f, 31.5f, 29.5f, 30.5f, 31.5f, 32.5f, 33.5f, 34.5f, 34.f, 35.f, 36.f, 38.5f, 39.5f, 40.5f, 41.5f, 42.5f, 43.5f, 43.f, 44.f, 45.f, 43.f, 44.f, 45.f, 46.f, 47.f, 48.f, 47.5f, 48.5f, 49.5f,
                                                           61.f, 62.f, 63.f, 64.f, 65.f, 66.f, 65.5f, 66.5f, 67.5f, 65.5f, 66.5f, 67.5f, 68.5f, 69.5f, 70.5f, 70.f, 71.f, 72.f, 74.5f, 75.5f, 76.5f, 77.5f, 78.5f, 79.5f, 79.f, 80.f, 81.f, 79.f, 80.f, 81.f, 82.f, 83.f, 84.f, 83.5f, 84.5f, 85.5f,
                                                           79.f, 80.f, 81.f, 82.f, 83.f, 84.f, 83.5f, 84.5f, 85.5f, 83.5f, 84.5f, 85.5f, 86.5f, 87.5f, 88.5f, 88.f, 89.f, 90.f, 92.5f, 93.5f, 94.5f, 95.5f, 96.5f, 97.5f, 97.f, 98.f, 99.f, 97.f, 98.f, 99.f, 100.f, 101.f, 102.f, 101.5f, 102.5f, 103.5f,
                                                          133.f, 134.f, 135.f, 136.f, 137.f, 138.f, 137.5f, 138.5f, 139.5f, 137.5f, 138.5f, 139.5f, 140.5f, 141.5f, 142.5f, 142.f, 143.f, 144.f, 146.5f, 147.5f, 148.5f, 149.5f, 150.5f, 151.5f, 151.f, 152.f, 153.f, 151.f, 152.f, 153.f, 154.f, 155.f, 156.f, 155.5f, 156.5f, 157.5f,
                                                          169.f, 170.f, 171.f, 172.f, 173.f, 174.f, 173.5f, 174.5f, 175.5f, 173.5f, 174.5f, 175.5f, 176.5f, 177.5f, 178.5f, 178.f, 179.f, 180.f, 182.5f, 183.5f, 184.5f, 185.5f, 186.5f, 187.5f, 187.f, 188.f, 189.f, 187.f, 188.f, 189.f, 190.f, 191.f, 192.f, 191.5f, 192.5f, 193.5f,
                                                          187.f, 188.f, 189.f, 190.f, 191.f, 192.f, 191.5f, 192.5f, 193.5f, 191.5f, 192.5f, 193.5f, 194.5f, 195.5f, 196.5f, 196.f, 197.f, 198.f, 200.5f, 201.5f, 202.5f, 203.5f, 204.5f, 205.5f, 205.f, 206.f, 207.f, 205.f, 206.f, 207.f, 208.f, 209.f, 210.f, 209.5f, 210.5f, 211.5f});
    input.linspace(1.);

    nd4j::ops::avgpool3dnew op;
    auto results = op.execute({&input}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW,  paddingMode, 0, 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(TypedConvolutionTests2, avgpool3d_test3) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 expected = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, iC}, {  29.5f, 30.5f, 31.5f, 32.5f, 33.5f, 34.5f, 38.5f, 39.5f, 40.5f, 41.5f, 42.5f, 43.5f, 65.5f, 66.5f, 67.5f, 68.5f, 69.5f, 70.5f,
                                                           74.5f, 75.5f, 76.5f, 77.5f, 78.5f, 79.5f, 137.5f, 138.5f, 139.5f, 140.5f, 141.5f, 142.5f, 146.5f, 147.5f, 148.5f, 149.5f, 150.5f, 151.5f,
                                                          173.5f, 174.5f, 175.5f, 176.5f, 177.5f, 178.5f, 182.5f, 183.5f, 184.5f, 185.5f, 186.5f, 187.5f});
    input.linspace(1.);

    nd4j::ops::avgpool3dnew op;
    auto results = op.execute({&input}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW,  paddingMode, 1, 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(TypedConvolutionTests2, avgpool3d_test4) {

    int bS=2, iD=3,iH=4,iW=3,  iC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=1,pH=1,pW=1, dD=1,dH=1,dW=1;
    int oD=4,oH=4,oW=4;
    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 expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, oD, oH, oW},{0.416667f, 1.00f, 1.333333f, 0.75f, 1.00f, 2.25f, 2.75f, 1.50f, 1.75f, 3.75f, 4.25f, 2.25f, 1.416667f, 3.00f, 3.333333f, 1.75f, 2.833333f, 6.00f, 6.666667f, 3.50f, 5.00f, 10.50f, 11.50f, 6.00f, 6.50f,
                                                        13.50f, 14.50f, 7.50f, 4.833333f, 10.00f, 10.666667f, 5.50f, 6.833333f, 14.00f, 14.666667f, 7.50f, 11.00f, 22.50f, 23.50f, 12.00f, 12.50f, 25.50f, 26.50f, 13.50f, 8.833333f, 18.00f, 18.666666f, 9.50f,
                                                        4.416667f, 9.00f, 9.333333f, 4.75f, 7.00f, 14.25f, 14.75f, 7.50f, 7.75f, 15.75f, 16.25f, 8.25f, 5.416667f, 11.00f, 11.333333f, 5.75f, 6.416667f, 13.00f, 13.333333f, 6.75f, 10.00f, 20.25f, 20.75f,
                                                        10.50f, 10.75f, 21.75f, 22.25f, 11.25f, 7.416667f, 15.00f, 15.333333f, 7.75f, 14.833333f, 30.00f, 30.666666f, 15.50f, 23.00f, 46.50f, 47.50f, 24.00f, 24.50f, 49.50f, 50.50f, 25.50f, 16.833334f,
                                                        34.00f, 34.666668f, 17.50f, 18.833334f, 38.00f, 38.666668f, 19.50f, 29.00f, 58.50f, 59.50f, 30.00f, 30.50f, 61.50f, 62.50f, 31.50f, 20.833334f, 42.00f, 42.666668f, 21.50f, 10.416667f, 21.00f,
                                                        21.333334f, 10.75f, 16.00f, 32.25f, 32.75f, 16.50f, 16.75f, 33.75f, 34.25f, 17.25f, 11.416667f, 23.00f, 23.333334f, 11.75f, 12.416667f, 25.00f, 25.333334f, 12.75f, 19.00f, 38.25f, 38.75f, 19.50f,
                                                        19.75f, 39.75f, 40.25f, 20.25f, 13.416667f, 27.00f, 27.333334f, 13.75f, 26.833334f, 54.00f, 54.666668f, 27.50f, 41.00f, 82.50f, 83.50f, 42.00f, 42.50f, 85.50f, 86.50f, 43.50f, 28.833334f, 58.00f,
                                                        58.666668f, 29.50f, 30.833334f, 62.00f, 62.666668f, 31.50f, 47.00f, 94.50f, 95.50f, 48.00f, 48.50f, 97.50f, 98.50f, 49.50f, 32.833332f, 66.00f, 66.666664f, 33.50f, 16.416666f, 33.00f, 33.333332f,
                                                        16.75f, 25.00f, 50.25f, 50.75f, 25.50f, 25.75f, 51.75f, 52.25f, 26.25f, 17.416666f, 35.00f, 35.333332f, 17.75f, 18.416666f, 37.00f, 37.333332f, 18.75f, 28.00f, 56.25f, 56.75f, 28.50f, 28.75f,
                                                        57.75f, 58.25f, 29.25f, 19.416666f, 39.00f, 39.333332f, 19.75f, 38.833332f, 78.00f, 78.666664f, 39.50f, 59.00f, 118.50f, 119.50f, 60.00f, 60.50f, 121.50f, 122.50f, 61.50f, 40.833332f, 82.00f,
                                                        82.666664f, 41.50f, 42.833332f, 86.00f, 86.666664f, 43.50f, 65.00f, 130.50f, 131.50f, 66.00f, 66.50f, 133.50f, 134.50f, 67.50f, 44.833332f, 90.00f, 90.666664f, 45.50f, 22.416666f, 45.00f,
                                                        45.333332f, 22.75f, 34.00f, 68.25f, 68.75f, 34.50f, 34.75f, 69.75f, 70.25f, 35.25f, 23.416666f, 47.00f, 47.333332f, 23.75f, 24.416666f, 49.00f, 49.333332f, 24.75f, 37.00f, 74.25f, 74.75f,
                                                        37.50f, 37.75f, 75.75f, 76.25f, 38.25f, 25.416666f, 51.00f, 51.333332f, 25.75f, 50.833332f, 102.00f, 102.666664f, 51.50f, 77.00f, 154.50f, 155.50f, 78.00f, 78.50f, 157.50f, 158.50f, 79.50f,
                                                        52.833332f, 106.00f, 106.666664f, 53.50f, 54.833332f, 110.00f, 110.666664f, 55.50f, 83.00f, 166.50f, 167.50f, 84.00f, 84.50f, 169.50f, 170.50f, 85.50f, 56.833332f, 114.00f, 114.666664f,
                                                        57.50f, 28.416666f, 57.00f, 57.333332f, 28.75f, 43.00f, 86.25f, 86.75f, 43.50f, 43.75f, 87.75f, 88.25f, 44.25f, 29.416666f, 59.00f, 59.333332f, 29.75f, 30.416666f, 61.00f, 61.333332f, 30.75f,
                                                        46.00f, 92.25f, 92.75f, 46.50f, 46.75f, 93.75f, 94.25f, 47.25f, 31.416666f, 63.00f, 63.333332f, 31.75f, 62.833332f, 126.00f, 126.666664f, 63.50f, 95.00f, 190.50f, 191.50f, 96.00f, 96.50f,
                                                        193.50f, 194.50f, 97.50f, 64.833336f, 130.00f, 130.666672f, 65.50f, 66.833336f, 134.00f, 134.666672f, 67.50f, 101.00f, 202.50f, 203.50f, 102.00f, 102.50f, 205.50f, 206.50f, 103.50f,
                                                        68.833336f, 138.00f, 138.666672f, 69.50f, 34.416668f, 69.00f, 69.333336f, 34.75f, 52.00f, 104.25f, 104.75f, 52.50f, 52.75f, 105.75f, 106.25f, 53.25f, 35.416668f, 71.00f, 71.333336f, 35.75f});
    input.linspace(1.);

    nd4j::ops::avgpool3dnew op;
    auto results = op.execute({&input}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW,  paddingMode, 1, 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(TypedConvolutionTests2, maxpool3d_test1) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, oD, oH, oW}, {20.f, 21.f, 23.f, 24.f, 32.f, 33.f, 35.f, 36.f, 56.f, 57.f, 59.f, 60.f, 68.f, 69.f, 71.f, 72.f, 92.f, 93.f, 95.f, 96.f, 104.f, 105.f, 107.f, 108.f,
                                                         128.f, 129.f, 131.f, 132.f, 140.f, 141.f, 143.f, 144.f, 164.f, 165.f, 167.f, 168.f, 176.f, 177.f, 179.f, 180.f, 200.f, 201.f, 203.f, 204.f, 212.f, 213.f, 215.f, 216.f});
    input.linspace(1.);

    nd4j::ops::maxpool3dnew op;
    auto results = op.execute({&input}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW, dD,dH,dW, paddingMode, 1, 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(TypedConvolutionTests2, maxpool3d_test2) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 expected = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, iC}, { 49.f, 50.f, 51.f, 52.f, 53.f, 54.f, 52.f, 53.f, 54.f, 58.f, 59.f, 60.f, 61.f, 62.f, 63.f, 61.f, 62.f, 63.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, 70.f, 71.f, 72.f, 70.f, 71.f, 72.f,
                                                          85.f, 86.f, 87.f, 88.f, 89.f, 90.f, 88.f, 89.f, 90.f, 94.f, 95.f, 96.f, 97.f, 98.f, 99.f, 97.f, 98.f, 99.f, 103.f, 104.f, 105.f, 106.f, 107.f, 108.f, 106.f, 107.f, 108.f, 103.f, 104.f, 105.f, 106.f, 107.f, 108.f, 106.f, 107.f, 108.f,
                                                          85.f, 86.f, 87.f, 88.f, 89.f, 90.f, 88.f, 89.f, 90.f, 94.f, 95.f, 96.f, 97.f, 98.f, 99.f, 97.f, 98.f, 99.f, 103.f, 104.f, 105.f, 106.f, 107.f, 108.f, 106.f, 107.f, 108.f, 103.f, 104.f, 105.f, 106.f, 107.f, 108.f, 106.f, 107.f, 108.f,
                                                         157.f, 158.f, 159.f, 160.f, 161.f, 162.f, 160.f, 161.f, 162.f, 166.f, 167.f, 168.f, 169.f, 170.f, 171.f, 169.f, 170.f, 171.f, 175.f, 176.f, 177.f, 178.f, 179.f, 180.f, 178.f, 179.f, 180.f, 175.f, 176.f, 177.f, 178.f, 179.f, 180.f, 178.f, 179.f, 180.f,
                                                         193.f, 194.f, 195.f, 196.f, 197.f, 198.f, 196.f, 197.f, 198.f, 202.f, 203.f, 204.f, 205.f, 206.f, 207.f, 205.f, 206.f, 207.f, 211.f, 212.f, 213.f, 214.f, 215.f, 216.f, 214.f, 215.f, 216.f, 211.f, 212.f, 213.f, 214.f, 215.f, 216.f, 214.f, 215.f, 216.f,
                                                         193.f, 194.f, 195.f, 196.f, 197.f, 198.f, 196.f, 197.f, 198.f, 202.f, 203.f, 204.f, 205.f, 206.f, 207.f, 205.f, 206.f, 207.f, 211.f, 212.f, 213.f, 214.f, 215.f, 216.f, 214.f, 215.f, 216.f, 211.f, 212.f, 213.f, 214.f, 215.f, 216.f, 214.f, 215.f, 216.f});
    input.linspace(1.);

    nd4j::ops::maxpool3dnew op;
    auto results = op.execute({&input}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, 1, 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(TypedConvolutionTests2, maxpool3d_test3) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 expected = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, iC}, {58.f, 59.f, 60.f, 61.f, 62.f, 63.f, 67.f, 68.f, 69.f, 70.f, 71.f, 72.f, 94.f, 95.f, 96.f, 97.f, 98.f, 99.f, 103.f, 104.f, 105.f, 106.f, 107.f, 108.f,
                                                         166.f, 167.f, 168.f, 169.f, 170.f, 171.f, 175.f, 176.f, 177.f, 178.f, 179.f, 180.f, 202.f, 203.f, 204.f, 205.f, 206.f, 207.f, 211.f, 212.f, 213.f, 214.f, 215.f, 216.f});
    input.linspace(1.);

    nd4j::ops::maxpool3dnew op;
    auto results = op.execute({&input}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW,  paddingMode, 1, 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(TypedConvolutionTests2, maxpool3d_test4) {

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

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, oD, oH, oW},{  4.f, 5.f, 6.f, 6.f, 7.f, 8.f, 9.f, 9.f, 10.f, 11.f, 12.f, 12.f, 10.f, 11.f, 12.f, 12.f, 16.f, 17.f, 18.f, 18.f, 19.f, 20.f, 21.f, 21.f, 22.f, 23.f, 24.f, 24.f, 22.f, 23.f, 24.f, 24.f, 28.f, 29.f, 30.f, 30.f, 31.f, 32.f, 33.f, 33.f, 34.f, 35.f, 36.f, 36.f, 34.f, 35.f, 36.f, 36.f,
                                                         28.f, 29.f, 30.f, 30.f, 31.f, 32.f, 33.f, 33.f, 34.f, 35.f, 36.f, 36.f, 34.f, 35.f, 36.f, 36.f, 40.f, 41.f, 42.f, 42.f, 43.f, 44.f, 45.f, 45.f, 46.f, 47.f, 48.f, 48.f, 46.f, 47.f, 48.f, 48.f, 52.f, 53.f, 54.f, 54.f, 55.f, 56.f, 57.f, 57.f, 58.f, 59.f, 60.f, 60.f, 58.f, 59.f, 60.f, 60.f,
                                                         64.f, 65.f, 66.f, 66.f, 67.f, 68.f, 69.f, 69.f, 70.f, 71.f, 72.f, 72.f, 70.f, 71.f, 72.f, 72.f, 64.f, 65.f, 66.f, 66.f, 67.f, 68.f, 69.f, 69.f, 70.f, 71.f, 72.f, 72.f, 70.f, 71.f, 72.f, 72.f, 76.f, 77.f, 78.f, 78.f, 79.f, 80.f, 81.f, 81.f, 82.f, 83.f, 84.f, 84.f, 82.f, 83.f, 84.f, 84.f,
                                                         88.f, 89.f, 90.f, 90.f, 91.f, 92.f, 93.f, 93.f, 94.f, 95.f, 96.f, 96.f, 94.f, 95.f, 96.f, 96.f, 100.f, 101.f, 102.f, 102.f, 103.f, 104.f, 105.f, 105.f, 106.f, 107.f, 108.f, 108.f, 106.f, 107.f, 108.f, 108.f, 100.f, 101.f, 102.f, 102.f, 103.f, 104.f, 105.f, 105.f, 106.f, 107.f, 108.f, 108.f, 106.f, 107.f, 108.f, 108.f,
                                                        112.f, 113.f, 114.f, 114.f, 115.f, 116.f, 117.f, 117.f, 118.f, 119.f, 120.f, 120.f, 118.f, 119.f, 120.f, 120.f, 124.f, 125.f, 126.f, 126.f, 127.f, 128.f, 129.f, 129.f, 130.f, 131.f, 132.f, 132.f, 130.f, 131.f, 132.f, 132.f, 136.f, 137.f, 138.f, 138.f, 139.f, 140.f, 141.f, 141.f, 142.f, 143.f, 144.f, 144.f, 142.f, 143.f, 144.f, 144.f,
                                                        136.f, 137.f, 138.f, 138.f, 139.f, 140.f, 141.f, 141.f, 142.f, 143.f, 144.f, 144.f, 142.f, 143.f, 144.f, 144.f, 148.f, 149.f, 150.f, 150.f, 151.f, 152.f, 153.f, 153.f, 154.f, 155.f, 156.f, 156.f, 154.f, 155.f, 156.f, 156.f, 160.f, 161.f, 162.f, 162.f, 163.f, 164.f, 165.f, 165.f, 166.f, 167.f, 168.f, 168.f, 166.f, 167.f, 168.f, 168.f,
                                                        172.f, 173.f, 174.f, 174.f, 175.f, 176.f, 177.f, 177.f, 178.f, 179.f, 180.f, 180.f, 178.f, 179.f, 180.f, 180.f, 172.f, 173.f, 174.f, 174.f, 175.f, 176.f, 177.f, 177.f, 178.f, 179.f, 180.f, 180.f, 178.f, 179.f, 180.f, 180.f, 184.f, 185.f, 186.f, 186.f, 187.f, 188.f, 189.f, 189.f, 190.f, 191.f, 192.f, 192.f, 190.f, 191.f, 192.f, 192.f,
                                                        196.f, 197.f, 198.f, 198.f, 199.f, 200.f, 201.f, 201.f, 202.f, 203.f, 204.f, 204.f, 202.f, 203.f, 204.f, 204.f, 208.f, 209.f, 210.f, 210.f, 211.f, 212.f, 213.f, 213.f, 214.f, 215.f, 216.f, 216.f, 214.f, 215.f, 216.f, 216.f, 208.f, 209.f, 210.f, 210.f, 211.f, 212.f, 213.f, 213.f, 214.f, 215.f, 216.f, 216.f, 214.f, 215.f, 216.f, 216.f});
    input.linspace(1.);

    nd4j::ops::maxpool3dnew op;
    auto results = op.execute({&input}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW, dD,dH,dW, paddingMode, 1, 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(TypedConvolutionTests2, avgpool3d_bp_test1) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oD, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW}, {0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.666667f, 1.333333f, 0.666667f, 0.666667f, 1.333333f, 0.666667f, 0.333333f, 0.666667f, 0.333333f,
                                                         0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f, 0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f,
                                                         0.333333f, 0.666667f, 0.333333f, 0.666667f, 1.333333f, 0.666667f, 0.666667f, 1.333333f, 0.666667f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f,
                                                         0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.666667f, 1.333333f, 0.666667f, 0.666667f, 1.333333f, 0.666667f, 0.333333f, 0.666667f, 0.333333f,
                                                         0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f, 0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f,
                                                         0.333333f, 0.666667f, 0.333333f, 0.666667f, 1.333333f, 0.666667f, 0.666667f, 1.333333f, 0.666667f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f,
                                                         0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.666667f, 1.333333f, 0.666667f, 0.666667f, 1.333333f, 0.666667f, 0.333333f, 0.666667f, 0.333333f,
                                                         0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f, 0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f,
                                                         0.333333f, 0.666667f, 0.333333f, 0.666667f, 1.333333f, 0.666667f, 0.666667f, 1.333333f, 0.666667f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f, 0.333333f, 0.666667f, 0.333333f, 0.333333f, 0.666667f, 0.333333f, 0.166667f, 0.333333f, 0.166667f});
    input.linspace(1.);
    gradO = 2.;

    nd4j::ops::avgpool3dnew_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW,  paddingMode, 1, 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(TypedConvolutionTests2, avgpool3d_bp_test2) {

    int bS=2, iD=3,iH=4,iW=3,  iC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=1,pH=1,pW=1, dD=1,dH=1,dW=1;
    int oD=4,oH=4,oW=4;
    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 gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oD, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW}, {1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f,
                                                         1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f,
                                                         1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f,
                                                         1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f,
                                                         1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f,
                                                         1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f,
                                                         1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f,
                                                         1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f,
                                                         1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 1.333333f, 2.f, 2.f, 2.f, 2.f, 2.f, 2.f, 1.333333f, 1.333333f, 1.333333f});
    input.linspace(1.);
    gradO = 2.;

    nd4j::ops::avgpool3dnew_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW,  paddingMode, 1, dataFormat});
    auto output = results->at(0);

    // output->printBuffer();

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

    delete results;
}


//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, avgpool3d_bp_test3) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, iC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC}, {0.41667f, 0.41667f, 0.41667f, 0.83333f, 0.83333f, 0.83333f, 1.25f, 1.25f, 1.25f, 0.58333f, 0.58333f, 0.58333f, 1.16667f, 1.16667f, 1.16667f, 1.75f, 1.75f, 1.75f, 0.58333f, 0.58333f, 0.58333f, 1.16667f, 1.16667f, 1.16667f, 1.75f, 1.75f, 1.75f,
                                                         0.41667f, 0.41667f, 0.41667f, 0.83333f, 0.83333f, 0.83333f, 1.25f, 1.25f, 1.25f, 0.83333f, 0.83333f, 0.83333f, 1.66667f, 1.66667f, 1.66667f, 2.5f, 2.5f, 2.5f, 1.16667f, 1.16667f, 1.16667f, 2.33333f, 2.33333f, 2.33333f, 3.5f, 3.5f, 3.5f,
                                                         1.16667f, 1.16667f, 1.16667f, 2.33333f, 2.33333f, 2.33333f, 3.5f, 3.5f, 3.5f, 0.83333f, 0.83333f, 0.83333f, 1.66667f, 1.66667f, 1.66667f, 2.5f, 2.5f, 2.5f, 1.25f, 1.25f, 1.25f, 2.5f, 2.5f, 2.5f, 3.75f, 3.75f, 3.75f,
                                                         1.75f, 1.75f, 1.75f, 3.5f, 3.5f, 3.5f, 5.25f, 5.25f, 5.25f, 1.75f, 1.75f, 1.75f, 3.5f, 3.5f, 3.5f, 5.25f, 5.25f, 5.25f, 1.25f, 1.25f, 1.25f, 2.5f, 2.5f, 2.5f, 3.75f, 3.75f, 3.75f,
                                                         0.41667f, 0.41667f, 0.41667f, 0.83333f, 0.83333f, 0.83333f, 1.25f, 1.25f, 1.25f, 0.58333f, 0.58333f, 0.58333f, 1.16667f, 1.16667f, 1.16667f, 1.75f, 1.75f, 1.75f, 0.58333f, 0.58333f, 0.58333f, 1.16667f, 1.16667f, 1.16667f, 1.75f, 1.75f, 1.75f,
                                                         0.41667f, 0.41667f, 0.41667f, 0.83333f, 0.83333f, 0.83333f, 1.25f, 1.25f, 1.25f, 0.83333f, 0.83333f, 0.83333f, 1.66667f, 1.66667f, 1.66667f, 2.5f, 2.5f, 2.5f, 1.16667f, 1.16667f, 1.16667f, 2.33333f, 2.33333f, 2.33333f, 3.5f, 3.5f, 3.5f,
                                                         1.16667f, 1.16667f, 1.16667f, 2.33333f, 2.33333f, 2.33333f, 3.5f, 3.5f, 3.5f, 0.83333f, 0.83333f, 0.83333f, 1.66667f, 1.66667f, 1.66667f, 2.5f, 2.5f, 2.5f, 1.25f, 1.25f, 1.25f, 2.5f, 2.5f, 2.5f, 3.75f, 3.75f, 3.75f,
                                                         1.75f, 1.75f, 1.75f, 3.5f, 3.5f, 3.5f, 5.25f, 5.25f, 5.25f, 1.75f, 1.75f, 1.75f, 3.5f, 3.5f, 3.5f, 5.25f, 5.25f, 5.25f, 1.25f, 1.25f, 1.25f, 2.5f, 2.5f, 2.5f, 3.75f, 3.75f, 3.75f});
    input.linspace(1.);
    gradO = 2.;

    nd4j::ops::avgpool3dnew_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW,  paddingMode, 0, 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(TypedConvolutionTests2, avgpool3d_bp_test4) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 = 0;             // 1-SAME,  0-VALID
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, iC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC}, {0.16667f, 0.16667f, 0.16667f, 0.33333f, 0.33333f, 0.33333f, 0.5f, 0.5f, 0.5f, 0.33333f, 0.33333f, 0.33333f, 0.66667f, 0.66667f, 0.66667f, 1.f, 1.f, 1.f, 0.58333f, 0.58333f, 0.58333f, 1.16667f, 1.16667f, 1.16667f, 1.75f, 1.75f, 1.75f,
                                                         0.91667f, 0.91667f, 0.91667f, 1.83333f, 1.83333f, 1.83333f, 2.75f, 2.75f, 2.75f, 0.33333f, 0.33333f, 0.33333f, 0.66667f, 0.66667f, 0.66667f, 1.f, 1.f, 1.f, 0.66667f, 0.66667f, 0.66667f, 1.33333f, 1.33333f, 1.33333f, 2.f, 2.f, 2.f,
                                                         1.16667f, 1.16667f, 1.16667f, 2.33333f, 2.33333f, 2.33333f, 3.5f, 3.5f, 3.5f, 1.83333f, 1.83333f, 1.83333f, 3.66667f, 3.66667f, 3.66667f, 5.5f, 5.5f, 5.5f, 0.5f, 0.5f, 0.5f, 1.f, 1.f, 1.f, 1.5f, 1.5f, 1.5f,
                                                         1.f, 1.f, 1.f, 2.f, 2.f, 2.f, 3.f, 3.f, 3.f, 1.75f, 1.75f, 1.75f, 3.5f, 3.5f, 3.5f, 5.25f, 5.25f, 5.25f, 2.75f, 2.75f, 2.75f, 5.5f, 5.5f, 5.5f, 8.25f, 8.25f, 8.25f,
                                                         0.16667f, 0.16667f, 0.16667f, 0.33333f, 0.33333f, 0.33333f, 0.5f, 0.5f, 0.5f, 0.33333f, 0.33333f, 0.33333f, 0.66667f, 0.66667f, 0.66667f, 1.f, 1.f, 1.f, 0.58333f, 0.58333f, 0.58333f, 1.16667f, 1.16667f, 1.16667f, 1.75f, 1.75f, 1.75f,
                                                         0.91667f, 0.91667f, 0.91667f, 1.83333f, 1.83333f, 1.83333f, 2.75f, 2.75f, 2.75f, 0.33333f, 0.33333f, 0.33333f, 0.66667f, 0.66667f, 0.66667f, 1.f, 1.f, 1.f, 0.66667f, 0.66667f, 0.66667f, 1.33333f, 1.33333f, 1.33333f, 2.f, 2.f, 2.f,
                                                         1.16667f, 1.16667f, 1.16667f, 2.33333f, 2.33333f, 2.33333f, 3.5f, 3.5f, 3.5f, 1.83333f, 1.83333f, 1.83333f, 3.66667f, 3.66667f, 3.66667f, 5.5f, 5.5f, 5.5f, 0.5f, 0.5f, 0.5f, 1.f, 1.f, 1.f, 1.5f, 1.5f, 1.5f,
                                                         1.f, 1.f, 1.f, 2.f, 2.f, 2.f, 3.f, 3.f, 3.f, 1.75f, 1.75f, 1.75f, 3.5f, 3.5f, 3.5f, 5.25f, 5.25f, 5.25f, 2.75f, 2.75f, 2.75f, 5.5f, 5.5f, 5.5f, 8.25f, 8.25f, 8.25f});
    input.linspace(1.);
    gradO = 2.;

    nd4j::ops::avgpool3dnew_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW,  paddingMode, 0, 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(TypedConvolutionTests2, maxpool3d_bp_test1) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oD, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW}, {0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.1f, 0.2f, 0.f, 0.3f, 0.4f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.5f, 0.6f, 0.f, 0.7f, 0.8f,
                0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.9f, 1.f, 0.f, 1.1f, 1.2f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 1.3f, 1.4f, 0.f, 1.5f, 1.6f,
                0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 1.7f, 1.8f, 0.f, 1.9f, 2.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 2.1f, 2.2f, 0.f, 2.3f, 2.4f,
                0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 2.5f, 2.6f, 0.f, 2.7f, 2.8f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 2.9f, 3.f, 0.f, 3.1f, 3.2f,
                0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 3.3f, 3.4f, 0.f, 3.5f, 3.6f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 3.7f, 3.8f, 0.f, 3.9f, 4.f,
                0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 4.1f, 4.2f, 0.f, 4.3f, 4.4f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 4.5f, 4.6f, 0.f, 4.7f, 4.8f});

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

    nd4j::ops::maxpool3dnew_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW, dD,dH,dW, paddingMode, 1, 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(TypedConvolutionTests2, maxpool3d_bp_test2) {

    int bS=2, iD=3,iH=4,iW=3,  iC=3,  kD=2,kH=3,kW=2,  sD=1,sH=1,sW=1,  pD=1,pH=1,pW=1, dD=1,dH=1,dW=1;
    int oD=4,oH=4,oW=4;
    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 gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oD, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iD, iH, iW}, {0.000e+00f, 0.000e+00f, 0.000e+00f, 1.000e-01f, 2.000e-01f, 7.000e-01f, 5.000e-01f, 6.000e-01f, 1.500e+00f, 2.200e+00f, 2.400e+00f, 5.400e+00f, 0.000e+00f, 0.000e+00f, 0.000e+00f, 1.700e+00f, 1.800e+00f, 3.900e+00f, 2.100e+00f, 2.200e+00f, 4.700e+00f, 5.400e+00f, 5.600e+00f, 1.180e+01f,
                                                         0.000e+00f, 0.000e+00f, 0.000e+00f, 8.200e+00f, 8.400e+00f, 1.740e+01f, 9.000e+00f, 9.200e+00f, 1.900e+01f, 2.040e+01f, 2.080e+01f, 4.280e+01f, 0.000e+00f, 0.000e+00f, 0.000e+00f, 6.500e+00f, 6.600e+00f, 1.350e+01f, 6.900e+00f, 7.000e+00f, 1.430e+01f, 1.500e+01f, 1.520e+01f, 3.100e+01f,
                                                         0.000e+00f, 0.000e+00f, 0.000e+00f, 8.100e+00f, 8.200e+00f, 1.670e+01f, 8.500e+00f, 8.600e+00f, 1.750e+01f, 1.820e+01f, 1.840e+01f, 3.740e+01f, 0.000e+00f, 0.000e+00f, 0.000e+00f, 2.100e+01f, 2.120e+01f, 4.300e+01f, 2.180e+01f, 2.200e+01f, 4.460e+01f, 4.600e+01f, 4.640e+01f, 9.400e+01f,
                                                         0.000e+00f, 0.000e+00f, 0.000e+00f, 1.290e+01f, 1.300e+01f, 2.630e+01f, 1.330e+01f, 1.340e+01f, 2.710e+01f, 2.780e+01f, 2.800e+01f, 5.660e+01f, 0.000e+00f, 0.000e+00f, 0.000e+00f, 1.450e+01f, 1.460e+01f, 2.950e+01f, 1.490e+01f, 1.500e+01f, 3.030e+01f, 3.100e+01f, 3.120e+01f, 6.300e+01f,
                                                         0.000e+00f, 0.000e+00f, 0.000e+00f, 3.380e+01f, 3.400e+01f, 6.860e+01f, 3.460e+01f, 3.480e+01f, 7.020e+01f, 7.160e+01f, 7.200e+01f, 1.452e+02f, 0.000e+00f, 0.000e+00f, 0.000e+00f, 1.930e+01f, 1.940e+01f, 3.910e+01f, 1.970e+01f, 1.980e+01f, 3.990e+01f, 4.060e+01f, 4.080e+01f, 8.220e+01f,
                                                         0.000e+00f, 0.000e+00f, 0.000e+00f, 2.090e+01f, 2.100e+01f, 4.230e+01f, 2.130e+01f, 2.140e+01f, 4.310e+01f, 4.380e+01f, 4.400e+01f, 8.860e+01f, 0.000e+00f, 0.000e+00f, 0.000e+00f, 4.660e+01f, 4.680e+01f, 9.420e+01f, 4.740e+01f, 4.760e+01f, 9.580e+01f, 9.720e+01f, 9.760e+01f, 1.964e+02f,
                                                         0.000e+00f, 0.000e+00f, 0.000e+00f, 2.570e+01f, 2.580e+01f, 5.190e+01f, 2.610e+01f, 2.620e+01f, 5.270e+01f, 5.340e+01f, 5.360e+01f, 1.078e+02f, 0.000e+00f, 0.000e+00f, 0.000e+00f, 2.730e+01f, 2.740e+01f, 5.510e+01f, 2.770e+01f, 2.780e+01f, 5.590e+01f, 5.660e+01f, 5.680e+01f, 1.142e+02f,
                                                         0.000e+00f, 0.000e+00f, 0.000e+00f, 5.940e+01f, 5.960e+01f, 1.198e+02f, 6.020e+01f, 6.040e+01f, 1.214e+02f, 1.228e+02f, 1.232e+02f, 2.476e+02f, 0.000e+00f, 0.000e+00f, 0.000e+00f, 3.210e+01f, 3.220e+01f, 6.470e+01f, 3.250e+01f, 3.260e+01f, 6.550e+01f, 6.620e+01f, 6.640e+01f, 1.334e+02f,
                                                         0.000e+00f, 0.000e+00f, 0.000e+00f, 3.370e+01f, 3.380e+01f, 6.790e+01f, 3.410e+01f, 3.420e+01f, 6.870e+01f, 6.940e+01f, 6.960e+01f, 1.398e+02f, 0.000e+00f, 0.000e+00f, 0.000e+00f, 7.220e+01f, 7.240e+01f, 1.454e+02f, 7.300e+01f, 7.320e+01f, 1.470e+02f, 1.484e+02f, 1.488e+02f, 2.988e+02f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::maxpool3dnew_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW, dD,dH,dW, paddingMode, 1, 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(TypedConvolutionTests2, maxpool3d_bp_test3) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, iC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC}, { 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f,
                                                          0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.1f, 0.2f, 0.3f, 1.1f, 1.3f, 1.5f,
                                                          0.f, 0.f, 0.f, 1.f, 1.1f, 1.2f, 2.9f, 3.1f, 3.3f, 0.f, 0.f, 0.f, 4.7f, 4.9f, 5.1f, 11.2f, 11.6f, 12.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f,
                                                          0.f, 0.f, 0.f, 11.f, 11.2f, 11.4f, 23.8f, 24.2f, 24.6f, 0.f, 0.f, 0.f, 12.8f, 13.f, 13.2f, 27.4f, 27.8f, 28.2f, 0.f, 0.f, 0.f, 31.f, 31.4f, 31.8f, 65.6f, 66.39999f, 67.2f,
                                                          0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f,
                                                          0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 10.9f, 11.f, 11.1f, 22.7f, 22.9f, 23.1f,
                                                          0.f, 0.f, 0.f, 11.8f, 11.9f, 12.f, 24.5f, 24.7f, 24.9f, 0.f, 0.f, 0.f, 26.3f, 26.5f, 26.7f, 54.4f, 54.8f, 55.2f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f,
                                                          0.f, 0.f, 0.f, 32.6f, 32.8f, 33.f, 67.f, 67.4f, 67.8f, 0.f, 0.f, 0.f, 34.4f, 34.6f, 34.8f, 70.6f, 71.f, 71.4f, 0.f, 0.f, 0.f, 74.2f, 74.6f, 75.f, 152.f, 152.8f, 153.6f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::maxpool3dnew_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, 1, 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(TypedConvolutionTests2, maxpool3d_bp_test4) {

    int bS=2, iD=3,iH=4,iW=3,  iC=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 = 0;             // 1-SAME,  0-VALID
    int dataFormat  = 1;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oD, oH, oW, iC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iD, iH, iW, iC}, {0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f,
                                                         0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.1f, 0.2f, 0.3f, 1.1f, 1.3f, 1.5f, 0.f, 0.f, 0.f, 5.7f, 6.f, 6.3f,
                                                         14.1f, 14.7f, 15.3f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 11.f, 11.2f, 11.4f, 23.8f, 24.2f,
                                                         24.6f, 0.f, 0.f, 0.f, 43.8f, 44.4f, 45.f, 93.f, 94.2f, 95.4f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f,
                                                         0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f,
                                                         10.9f, 11.f, 11.1f, 22.7f, 22.9f, 23.1f, 0.f, 0.f, 0.f, 38.1f, 38.4f, 38.7f, 78.9f, 79.5f, 80.1f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f,
                                                         0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 32.6f, 32.8f, 33.f, 67.f, 67.4f, 67.8f, 0.f, 0.f, 0.f, 108.6f, 109.2f, 109.8f, 222.6f, 223.8f, 225.f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::maxpool3dnew_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW, dD,dH,dW,  paddingMode, 1, dataFormat});
    auto output = results->at(0);

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, maxpool2d_bp_1) {

    auto input = NDArrayFactory::create_<float>('c', {bS,iD,iH,iW});
    auto epsilon = NDArrayFactory::create_<float>('c', {bS,iD,oH,oW});
    auto exp     = NDArrayFactory::create<float>('c', {bS,iD,iH,iW});

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, input);
    variableSpace->putVariable(-2, epsilon);
    // variableSpace->putVariable(1, &z);

    auto block = new Context(1, variableSpace, false);
    block->fillInputs({-1});
    block->fillInputs({-2});
    std::vector<int>* argI = block->getIArguments();
    *argI = {kH,kW, sH,sW, pH,pW, dW,dH, 0, 0, 0};   // 0,1 - kernel Height/Width; 2,3 - stride Height/Width; 4,5 - pad Height/Width; 6,7 - dilation Height/Width; 8 - same mode;

    nd4j::ops::maxpool2d_bp bp;
    Nd4jStatus status = bp.execute(block);
    ASSERT_EQ(ND4J_STATUS_OK, status);

    auto result = variableSpace->getVariable(block->getNodeId())->getNDArray();
    ASSERT_TRUE(exp.isSameShape(result));

    delete variableSpace;
    delete block;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, maxpool2d_bp_2) {

    int bS=2, iD=1, iH=4,iW=4, oD=3, kH=2,kW=2, sH=1,sW=1, pH=0,pW=0, dH=1,dW=1;
    int oH = (iH - kH - (kH-1)*(dH-1) + 2*pH)/sH + 1;
    int oW = (iW - kW - (kW-1)*(dW-1) + 2*pW)/sW + 1;

    // TypeParam epsilonBuff[]  = {6., 7., 8., 10., 11., 12., 14., 15., 16., 22., 23., 24., 26., 27., 28., 30., 31., 32.};
    // TypeParam expectedBuff[] = {0., 0., 0., 0.,0., 6., 7., 8.,0.,10.,11.,12.,0.,14.,15.,16.,0., 0., 0., 0.,0.,22.,23.,24.,0.,26.,27.,28.,0.,30.,31.,32.};

    NDArray input('c', {bS,iD,iH,iW});
    NDArray epsilon('c', {bS,iD,oH,oW}, {6., 7., 8., 10., 11., 12., 14., 15., 16., 22., 23., 24., 26., 27., 28., 30., 31., 32.});
    NDArray expected('c', {bS,iD,iH,iW}, {0., 0., 0., 0.,0., 6., 7., 8.,0.,10.,11.,12.,0.,14.,15.,16.,0., 0., 0., 0.,0.,22.,23.,24.,0.,26.,27.,28.,0.,30.,31.,32.});


    input.linspace(1.);

    std::initializer_list<Nd4jLong> argI = {kH,kW, sH,sW, pH,pW, dW,dH, 0, 0, 0};   // 0,1 - kernel Height/Width; 2,3 - stride Height/Width; 4,5 - pad Height/Width; 6,7 - dilation Height/Width; 8 - same mode;

    nd4j::ops::maxpool2d_bp op;
    auto results = op.execute({&input, &epsilon}, {}, argI);
    auto output = results->at(0);

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, maxpool2d_bp_3) {

    int bS=2, iH=4,iW=3,  iC=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-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW}, {0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.1f, 0.2f, 0.f, 0.3f, 0.4f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.5f, 0.6f, 0.f, 0.7f, 0.8f,
                                                     0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.9f, 1.f, 0.f, 1.1f, 1.2f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 1.3f, 1.4f, 0.f, 1.5f, 1.6f,
                                                     0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 1.7f, 1.8f, 0.f, 1.9f, 2.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 2.1f, 2.2f, 0.f, 2.3f, 2.4f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::maxpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, 1, 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(TypedConvolutionTests2, maxpool2d_bp_4) {

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

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW}, {0.f, 0.f, 0.f, 0.1f, 0.2f, 0.7f, 0.5f, 0.6f, 1.5f, 2.2f, 2.4f, 5.4f, 0.f, 0.f, 0.f, 1.7f, 1.8f, 3.9f, 2.1f, 2.2f, 4.7f, 5.4f, 5.6f, 11.8f,
                                                     0.f, 0.f, 0.f, 3.3f, 3.4f, 7.1f, 3.7f, 3.8f, 7.9f, 8.6f, 8.8f, 18.2f, 0.f, 0.f, 0.f, 4.9f, 5.f, 10.3f, 5.3f, 5.4f, 11.1f, 11.8f, 12.f, 24.6f,
                                                     0.f, 0.f, 0.f, 6.5f, 6.6f, 13.5f, 6.9f, 7.f, 14.3f, 15.f, 15.2f, 31.f, 0.f, 0.f, 0.f, 8.1f, 8.2f, 16.7f, 8.5f, 8.6f, 17.5f, 18.2f, 18.4f, 37.4f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::maxpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, 1, 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(TypedConvolutionTests2, maxpool2d_bp_5) {

    int bS=2, iH=4,iW=3,  iC=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-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, iC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC}, {0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.1f, 0.2f, 0.3f, 1.1f, 1.3f, 1.5f, 0.f, 0.f, 0.f, 1.f, 1.1f, 1.2f, 2.9f, 3.1f, 3.3f,
                                                     0.f, 0.f, 0.f, 4.7f, 4.9f, 5.1f, 11.2f, 11.6f, 12.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 3.7f, 3.8f, 3.9f, 8.3f, 8.5f, 8.7f,
                                                     0.f, 0.f, 0.f, 4.6f, 4.7f, 4.8f, 10.1f, 10.3f, 10.5f, 0.f, 0.f, 0.f, 11.9f, 12.1f, 12.3f, 25.6f, 26.f, 26.4f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::maxpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, 1, 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(TypedConvolutionTests2, maxpool2d_bp_6) {

    int bS=2, iH=4,iW=3,  iC=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-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, iC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC}, {0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.1f, 0.2f, 0.3f, 0.4f, 0.5f, 0.6f,
                                                     0.f, 0.f, 0.f, 0.7f, 0.8f, 0.9f, 1.f, 1.1f, 1.2f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f,
                                                     0.f, 0.f, 0.f, 1.3f, 1.4f, 1.5f, 1.6f, 1.7f, 1.8f, 0.f, 0.f, 0.f, 1.9f, 2.f, 2.1f, 2.2f, 2.3f, 2.4f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::maxpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, 1, dataFormat});
    auto output = results->at(0);

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, maxpool2d_bp_7) {

    int bS=2, iH=56,iW=56,  iC=3,  kH=2,kW=2,  sH=2,sW=2,  pH=0,pW=0,  dH=1,dW=1;
    int       oH=28,oW=28;
    int paddingMode = 1;             // 1-SAME,  0-VALID
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<float16>('c', {bS, iC, iH, iW});
    auto gradO    = NDArrayFactory::create<float16>('c', {bS, iC, oH, oW});

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

    nd4j::ops::maxpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, 1, dataFormat});
    // auto output = results->at(0);

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, avgpool2d_bp_1) {

    auto input = NDArrayFactory::create_<float>('c', {bS,iD,iH,iW});
    auto epsilon = NDArrayFactory::create_<float>('c', {bS,iD,oH,oW});
    auto exp     = NDArrayFactory::create<float>('c', {bS,iD,iH,iW});

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, input);
    variableSpace->putVariable(-2, epsilon);
    // variableSpace->putVariable(1, &z);

    auto block = new Context(1, variableSpace, false);
    block->fillInputs({-1});
    block->fillInputs({-2});
    std::vector<int>* argI = block->getIArguments();
    *argI = {kH,kW, sH,sW, pH,pW, dW,dH, 0, 1, 0};   // 0,1 - kernel Height/Width; 2,3 - stride Height/Width; 4,5 - pad Height/Width; 6,7 - dilation Height/Width; 8 - same mode, 9 - extraParam0 (unnecessary for avg mode), 10 - data format

    nd4j::ops::avgpool2d_bp bp;
    Nd4jStatus status = bp.execute(block);
    ASSERT_EQ(ND4J_STATUS_OK, status);

    auto result = variableSpace->getVariable(block->getNodeId())->getNDArray();
    ASSERT_TRUE(exp.isSameShape(result));

    delete variableSpace;
    delete block;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, avgpool2d_bp_2) {

    int bS=2, iD=1, iH=4,iW=4, oD=3, kH=2,kW=2, sH=1,sW=1, pH=0,pW=0, dH=1,dW=1;
    int oH = (iH - kH - (kH-1)*(dH-1) + 2*pH)/sH + 1;
    int oW = (iW - kW - (kW-1)*(dW-1) + 2*pW)/sW + 1;

    // TypeParam epsilonBuff[] = {3.5 , 4.5 , 5.5, 7.5 , 8.5 , 9.5, 11.5, 12.5, 13.5, 19.5, 20.5, 21.5, 23.5, 24.5, 25.5, 27.5, 28.5, 29.5};
    // TypeParam expectedBuff[] = {0.875, 2., 2.5,1.375, 2.75 , 6., 7.,  3.75, 4.75 ,10., 11., 5.75, 2.875, 6., 6.5, 3.375, 4.875, 10.,10.5, 5.375, 10.75, 22.,23., 11.75, 12.75, 26.,27., 13.75, 6.875, 14.,14.5, 7.375};

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS,iD,iH,iW});
    auto epsilon  = NDArrayFactory::create<TypeParam>('c', {bS,iD,oH,oW}, {3.5f, 4.5f, 5.5f, 7.5f, 8.5f, 9.5f, 11.5f, 12.5f, 13.5f, 19.5f, 20.5f, 21.5f, 23.5f, 24.5f, 25.5f, 27.5f, 28.5f, 29.5f});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS,iD,iH,iW}, {0.875f, 2.f, 2.5f, 1.375f, 2.75f, 6.f, 7.f,  3.75f, 4.75f, 10.f, 11.f, 5.75f, 2.875f, 6.f, 6.5f, 3.375f, 4.875f, 10.f, 10.5f, 5.375f, 10.75f, 22.f, 23.f, 11.75f, 12.75f, 26.f, 27.f, 13.75f, 6.875f, 14.f, 14.5f, 7.375f});

    input.linspace(1.);

    std::initializer_list<Nd4jLong> argI = {kH,kW, sH,sW, pH,pW, dW,dH, 1, 1, 0};

    nd4j::ops::avgpool2d_bp op;
    auto results = op.execute({&input, &epsilon}, {}, argI);
    auto output = results->at(0);

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, avgpool2d_bp_3) {

    int bS=2, iH=4,iW=3,  iC=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-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW}, {0.016667f, 0.05f, 0.033333f, 0.066667f, 0.166667f, 0.1f, 0.066667f, 0.166667f, 0.1f, 0.05f, 0.116667f, 0.066667f,
                                                     0.083333f, 0.183333f, 0.1f, 0.2f, 0.433333f, 0.233333f, 0.2f, 0.433333f, 0.233333f, 0.116667f, 0.25f, 0.133333f,
                                                     0.15f, 0.316667f, 0.166667f, 0.333333f, 0.7f, 0.366667f, 0.333333f, 0.7f, 0.366667f, 0.183333f, 0.383333f, 0.2f,
                                                     0.216667f, 0.45f, 0.233333f, 0.466667f, 0.966667f, 0.5f, 0.466667f, 0.966667f, 0.5f, 0.25f, 0.516667f, 0.266667f,
                                                     0.283333f, 0.583333f, 0.3f, 0.6f, 1.233333f, 0.633333f, 0.6f, 1.233333f, 0.633333f, 0.316667f, 0.65f, 0.333333f,
                                                     0.35f, 0.716667f, 0.366667f, 0.733333f, 1.5f, 0.766667f, 0.733333f, 1.5f, 0.766667f, 0.383333f, 0.783333f, 0.4f     });
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::avgpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, 1, 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(TypedConvolutionTests2, avgpool2d_bp_4) {

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

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW}, {0.233333f, 0.3f, 0.366667f, 0.55f, 0.65f, 0.75f, 0.95f, 1.05f, 1.15f, 0.766667f, 0.833333f, 0.9f,
                                                     1.3f, 1.366667f, 1.433333f, 2.15f, 2.25f, 2.35f, 2.55f, 2.65f, 2.75f, 1.833333f, 1.9f, 1.966667f,
                                                     2.366667f, 2.433333f, 2.5f, 3.75f, 3.85f, 3.95f, 4.15f, 4.25f, 4.35f, 2.9f, 2.966667f, 3.033333f,
                                                     3.433333f, 3.5f, 3.566667f, 5.35f, 5.45f, 5.55f, 5.75f, 5.85f, 5.95f, 3.966667f, 4.033333f, 4.1f,
                                                     4.5f, 4.566667f, 4.633333f, 6.95f, 7.05f, 7.15f, 7.35f, 7.45f, 7.55f, 5.033333f, 5.1f, 5.166667f,
                                                     5.566667f, 5.633333f, 5.7f, 8.549999f, 8.65f, 8.75f, 8.95f, 9.05f, 9.150001f, 6.1f, 6.166667f, 6.233334f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::avgpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, 1, 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(TypedConvolutionTests2, avgpool2d_bp_5) {

    int bS=2, iH=4,iW=3,  iC=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-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, iC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC}, {0.19167f, 0.23333f, 0.275f, 0.50833f, 0.59167f, 0.675f, 1.2f, 1.325f, 1.45f, 0.50833f, 0.56667f, 0.625f, 1.19167f, 1.30833f, 1.425f, 2.4f, 2.575f, 2.75f,
                                                     1.18333f, 1.24167f, 1.3f, 2.54167f, 2.65833f, 2.775f, 4.425f, 4.6f, 4.775f, 1.01667f, 1.05833f, 1.1f, 2.15833f, 2.24167f, 2.325f, 3.675f, 3.8f, 3.925f,
                                                     1.69167f, 1.73333f, 1.775f, 3.50833f, 3.59167f, 3.675f, 5.7f, 5.825f, 5.95f, 2.60833f, 2.66667f, 2.725f, 5.39167f, 5.50833f, 5.625f, 8.7f, 8.875f, 9.05f,
                                                     3.28333f, 3.34167f, 3.4f, 6.74167f, 6.85833f, 6.975f, 10.725f, 10.9f, 11.075f, 2.51667f, 2.55833f, 2.6f, 5.15833f, 5.24167f, 5.325f, 8.175f, 8.3f, 8.425f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::avgpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, 0, 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(TypedConvolutionTests2, avgpool2d_bp_6) {

    int bS=2, iH=4,iW=3,  iC=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-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, oH, oW, iC});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iH, iW, iC}, {0.01667f, 0.03333f, 0.05f, 0.08333f, 0.11667f, 0.15f, 0.06667f, 0.08333f, 0.1f, 0.13333f, 0.16667f, 0.2f, 0.36667f, 0.43333f, 0.5f, 0.23333f, 0.26667f, 0.3f,
                                                     0.13333f, 0.16667f, 0.2f, 0.36667f, 0.43333f, 0.5f, 0.23333f, 0.26667f, 0.3f, 0.11667f, 0.13333f, 0.15f, 0.28333f, 0.31667f, 0.35f, 0.16667f, 0.18333f, 0.2f,
                                                     0.21667f, 0.23333f, 0.25f, 0.48333f, 0.51667f, 0.55f, 0.26667f, 0.28333f, 0.3f, 0.53333f, 0.56667f, 0.6f, 1.16667f, 1.23333f, 1.3f, 0.63333f, 0.66667f, 0.7f,
                                                     0.53333f, 0.56667f, 0.6f, 1.16667f, 1.23333f, 1.3f, 0.63333f, 0.66667f, 0.7f, 0.31667f, 0.33333f, 0.35f, 0.68333f, 0.71667f, 0.75f, 0.36667f, 0.38333f, 0.4f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::avgpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, 1, dataFormat});
    auto output = results->at(0);

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, pnormpool2d_bp_1) {

    auto input = NDArrayFactory::create_<float>('c', {bS,iD,iH,iW});
    auto epsilon = NDArrayFactory::create_<float>('c', {bS,iD,oH,oW});
    auto exp     = NDArrayFactory::create<float>('c', {bS,iD,iH,iW});

    auto variableSpace = new VariableSpace();
    variableSpace->putVariable(-1, input);
    variableSpace->putVariable(-2, epsilon);
    // variableSpace->putVariable(1, &z);

    auto block = new Context(1, variableSpace, false);
    block->fillInputs({-1});
    block->fillInputs({-2});
    auto argI = block->getIArguments();
    *argI = {kH,kW, sH,sW, pH,pW, dW,dH, 0, 3};   // 0,1 - kernel Height/Width; 2,3 - stride Height/Width; 4,5 - pad Height/Width; 6,7 - dilation Height/Width; 8 - same mode; 9 - divisor
    std::vector<double>* argT = block->getTArguments();
    *argT = {0.000001};

    nd4j::ops::pnormpool2d_bp bp;
    Nd4jStatus status = bp.execute(block);
    ASSERT_EQ(ND4J_STATUS_OK, status);

    auto result = variableSpace->getVariable(block->getNodeId())->getNDArray();
    ASSERT_TRUE(exp.isSameShape(result));

    delete variableSpace;
    delete block;
}

//////////////////////////////////////////////////////////////////////
TYPED_TEST(TypedConvolutionTests2, pnormpool2d_bp_2) {

    int bS=2, iH=4,iW=3,  iC=3,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int oH=2,oW=2;
    int pnorm = 3;
    double eps = 0.;

    int paddingMode = 0;             // 1-SAME,  0-VALID
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW}, {9.661570e-04f, 9.671602e-03f, 1.306569e-02f, 3.679184e-02f, 1.297220e-01f, 1.040181e-01f, 1.126750e-01f, 3.320884e-01f, 2.340406e-01f, 1.333333e-01f, 3.352886e-01f, 2.070211e-01f,
                                                     8.991618e-02f, 2.160601e-01f, 1.283173e-01f, 2.744226e-01f, 6.364498e-01f, 3.662123e-01f, 3.869788e-01f, 8.808994e-01f, 4.984556e-01f, 2.613189e-01f, 5.818475e-01f, 3.225517e-01f,
                                                     2.065654e-01f, 4.553546e-01f, 2.501175e-01f, 5.190718e-01f, 1.131343e+00f, 6.148388e-01f, 6.362602e-01f, 1.377521e+00f, 7.439550e-01f, 3.833026e-01f, 8.227519e-01f, 4.407146e-01f,
                                                     3.261206e-01f, 6.969233e-01f, 3.717564e-01f, 7.627507e-01f, 1.620991e+00f, 8.600952e-01f, 8.814538e-01f, 1.866888e+00f, 9.873542e-01f, 5.046682e-01f, 1.064004e+00f, 5.602558e-01f,
                                                     4.464697e-01f, 9.389536e-01f, 4.932274e-01f, 1.005908e+00f, 2.108550e+00f, 1.104095e+00f, 1.125322e+00f, 2.354009e+00f, 1.230180e+00f, 6.258913e-01f, 1.305581e+00f, 6.804127e-01f,
                                                     5.671396e-01f, 1.181128e+00f, 6.145977e-01f, 1.248783e+00f, 2.595083e+00f, 1.347494e+00f, 1.368600e+00f, 2.840157e+00f, 1.472778e+00f, 7.470673e-01f, 1.547362e+00f, 8.008900e-01f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::pnormpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {eps}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, pnorm, 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(TypedConvolutionTests2, pnormpool2d_bp_3) {

    int bS=2, iH=4,iW=3,  iC=3,  kH=3,kW=2,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int oH=2,oW=2;
    int pnorm = 2;
    double eps = 0.;

    int paddingMode = 0;             // 1-SAME,  0-VALID
    int dataFormat  = 0;             // 1-NDHWC, 0-NCDHW

    auto input    = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW});
    auto gradO    = NDArrayFactory::create<TypeParam>('c', {bS, iC, oH, oW});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS, iC, iH, iW}, {0.007931f, 0.042891f, 0.040544f, 0.09369f, 0.276841f, 0.191675f, 0.163957f, 0.442946f, 0.287512f, 0.154919f, 0.373153f, 0.221172f,
                                                     0.15901f, 0.365232f, 0.207846f, 0.428282f, 0.959455f, 0.534076f, 0.508585f, 1.128771f, 0.623089f, 0.319794f, 0.698063f, 0.379547f,
                                                     0.321068f, 0.692438f, 0.372316f, 0.757521f, 1.620323f, 0.864566f, 0.838684f, 1.787943f, 0.951023f, 0.483194f, 1.023434f, 0.541058f,
                                                     0.483937f, 1.019414f, 0.536145f, 1.085348f, 2.276996f, 1.192917f, 1.166749f, 2.443606f, 1.278126f, 0.646499f, 1.349361f, 0.703463f,
                                                     0.647021f, 1.346249f, 0.699745f, 1.412654f, 2.932174f, 1.520512f, 1.494153f, 3.098146f, 1.604985f, 0.809791f, 1.675544f, 0.866229f,
                                                     0.810192f, 1.673009f, 0.863237f, 1.739711f, 3.58665f, 1.847753f, 1.82126f, 3.752188f, 1.931741f, 0.973081f, 2.001861f, 1.029173f});
    input.linspace(1.);
    gradO.linspace(0.1, 0.1);

    nd4j::ops::pnormpool2d_bp op;
    auto results = op.execute({&input, &gradO}, {eps}, {kH,kW,  sH,sW,  pH,pW,  dH,dW,  paddingMode, pnorm, dataFormat});
    auto output = results->at(0);

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

    delete results;
}


//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, upsampling2d_bp_1) {

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

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

    auto expGradI = NDArrayFactory::create<float>('c', {bS, iC, iH, iW});
    expGradI = 4.;

    nd4j::ops::upsampling2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {isNCHW});
    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(ConvolutionTests2, upsampling2d_bp_2) {

    const int bS=1,  iH=2,iW=2,  iC=1;
    const int factorH=2, factorW=2;
    const int isNCHW = 0;                    // data format, default is NCHW

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

    auto expGradI = NDArrayFactory::create<float>('c', {bS, iH, iW, iC});
    expGradI = 4.;

    nd4j::ops::upsampling2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {isNCHW});
    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(ConvolutionTests2, upsampling2d_bp_3) {

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

    NDArray input('c', {bS, iC, iH, iW}, nd4j::DataType::FLOAT32);

    NDArray gradO('c', {bS, iC, 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}, nd4j::DataType::FLOAT32);

    NDArray expGradI('c', {bS, iC, iH, iW}, {2.4203868, 1.5216494, 2.1776323, 2.0290341, 0.772146, 1.5008594, 1.0523045, 1.3174672, 1.9263644,
                    1.090545, 1.9094483, 1.3611296, 2.1195147, 2.0659215, 1.0423062, 2.3405795, 1.9105877, 1.2203633}, nd4j::DataType::FLOAT32);

    nd4j::ops::upsampling2d_bp op;
    auto results = op.execute({&input, &gradO}, {}, {isNCHW});
    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(TypedConvolutionTests2, 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(ConvolutionTests2, 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<float>('c', {bS, iH, iW, iC});
    auto weights  = NDArrayFactory::create<float>('c', {kH, kW, iC, mC});


    auto expOutput = NDArrayFactory::create<float>('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(ConvolutionTests2, 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<float>('c', {bS, iC, iH, iW});
    auto weights  = NDArrayFactory::create<float>('c', {mC, iC, kH, kW});
    auto biases   = NDArrayFactory::create<float>('c', {iC*mC}, {1.f,2.f,3.f,4.f});

    NDArray expOutput('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}, nd4j::DataType::FLOAT32);

    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);

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

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, 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(ConvolutionTests2, 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<float>('c', {bS, iH, iW, iC});
    auto weights = NDArrayFactory::create<float>('c', {kH, kW, iC, mC});

    NDArray expOutput('c', {bS, oH, oW, oC}, {10., 12., 14., 16., 8., 9., 22., 24., 26., 28., 14., 15., 14., 15., 16., 17., 8.5, 9.}, nd4j::DataType::FLOAT32);

    input.linspace(1.);
    weights = 0.5;

    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(ConvolutionTests2, 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::FLOAT32);
    NDArray weights('c', {kH, kW, iC, mC}, nd4j::DataType::FLOAT32);

    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.}, nd4j::DataType::FLOAT32);
    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(ConvolutionTests2, 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}, nd4j::DataType::FLOAT32);
    NDArray weights('c', {kH, kW, iC, mC}, {0.1308445781469345, 0.6442840099334717, 0.5698848366737366, 0.19896849989891052}, nd4j::DataType::FLOAT32);
    NDArray biases('c', {1,iC*mC}, {0.6123566627502441, 0.37637925148010254, 0.17464971542358398, 0.4270855486392975}, nd4j::DataType::FLOAT32);

    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::DataType::FLOAT32);


    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(ConvolutionTests2, depthwise_conv2d_8) {

    int bS=1, iH=10,iW=10,  iC=8,mC=1,  kH=3,kW=3,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oC=iC*mC;
    int       oH=10,oW=10;
    int paddingMode = 1;             // 1-SAME, 0-VALID;
    int dataFormat  = 1;             // 1-NHWC, 0-NCHW

    NDArray input('c', {bS, iH, iW, iC}, nd4j::DataType::FLOAT32);
    NDArray weights('c', {kH, kW, iC, mC}, nd4j::DataType::FLOAT32);

    NDArray expOutput('c', {bS, oH, oW, oC}, {-42.879997, -43.959999, -44.959999, -45.879997, -46.720005, -47.480003, -48.160000, -48.760002, -43.519997, -45.139999, -46.639996, -48.020000, -49.280003, -50.419998, -51.440006, -52.340000, -31.999998, -33.139999, -34.160000, -35.060001, -35.840004, -36.500004, -37.039997, -37.459999, -20.480000,
        -21.139997, -21.680000, -22.100000, -22.399998, -22.579998, -22.639996, -22.580002, -8.960000, -9.139998, -9.200002, -9.140001, -8.960001, -8.660000, -8.240002, -7.700001, 2.560000, 2.860002, 3.279998, 3.820000, 4.480001, 5.260000, 6.160001, 7.180000, 14.080000, 14.860000, 15.759998, 16.779999, 17.920002, 19.180000, 20.560001, 22.059998,
        25.600000, 26.860001, 28.239998, 29.739998, 31.360001, 33.099998, 34.959999, 36.939999, 37.119999, 38.860001, 40.720001, 42.699997, 44.800003, 47.020000, 49.360001, 51.820000, 26.239998, 27.400002, 28.639999, 29.959999, 31.360001, 32.840000, 34.400002, 36.040001, 62.400002, 62.459999, 62.639999, 62.940002, 63.360001, 63.900002, 64.559998,
        65.340004, 106.080002, 106.169998, 106.440002, 106.889999, 107.519997, 108.330002, 109.320000, 110.490005, 114.720001, 115.529999, 116.520004, 117.690002, 119.040009, 120.570000, 122.279999, 124.169998, 123.359985, 124.889999, 126.599998, 128.490005, 130.559998, 132.809998, 135.240005, 137.850006, 132.000000, 134.250000, 136.679993,
        139.290009, 142.080002, 145.049988, 148.199997, 151.529999, 140.639999, 143.610001, 146.760010, 150.089996, 153.600006, 157.290009, 161.160004, 165.209991, 149.279999, 152.970001, 156.839996, 160.889999, 165.120010, 169.529999, 174.119995, 178.889999, 157.919998, 162.330002, 166.919983, 171.690002, 176.639999, 181.769989, 187.079987,
        192.570007, 166.559998, 171.690002, 177.000000, 182.489990, 188.160004, 194.010010, 200.040009, 206.250000, 100.799995, 104.220001, 107.760002, 111.419998, 115.200005, 119.099998, 123.120003, 127.260010, 139.200012, 144.059998, 149.040009, 154.139999, 159.360001, 164.699997, 170.160004, 175.739990, 192.479996, 199.770020, 207.239990,
        214.889999, 222.720001, 230.730011, 238.919998, 247.290009, 201.119995, 209.129990, 217.319992, 225.690002, 234.240005, 242.970001, 251.880005, 260.970001, 209.760010, 218.489990, 227.399994, 236.490005, 245.760010, 255.209991, 264.839996, 274.649994, 218.399994, 227.850006, 237.479996, 247.289993, 257.279999, 267.449982, 277.799988,
        288.330017, 227.040009, 237.209991, 247.559998, 258.089996, 268.800018, 279.690002, 290.760010, 302.010010, 235.679993, 246.570007, 257.639984, 268.889984, 280.320007, 291.929993, 303.720001, 315.690002, 244.320007, 255.929993, 267.720001, 279.690002, 291.839996, 304.169983, 316.679993, 329.369995, 252.959991, 265.290009, 277.799988,
        290.489990, 303.359985, 316.410004, 329.640015, 343.050018, 139.199997, 147.419998, 155.760010, 164.220001, 172.799988, 181.500000, 190.319992, 199.260010, 216.000000, 225.660004, 235.440002, 245.339996, 255.360016, 265.500000, 275.760010, 286.140015, 278.880005, 293.369995, 308.040009, 322.889984, 337.920013, 353.129974, 368.519989,
        384.090027, 287.520020, 302.730011, 318.119995, 333.690002, 349.440002, 365.369995, 381.479980, 397.770020, 296.160004, 312.089996, 328.199982, 344.489990, 360.960022, 377.609985, 394.440002, 411.449982, 304.799988, 321.450012, 338.280029, 355.289978, 372.480011, 389.850006, 407.399994, 425.130005, 313.440002, 330.809998, 348.359985, 366.089996, 384.000000, 402.090027, 420.359985, 438.809998, 322.079987, 340.169983, 358.440002, 376.889984, 395.520020, 414.329987, 433.320007, 452.489990, 330.720001, 349.530029, 368.520020, 387.690002, 407.039978, 426.570007, 446.279999, 466.170013, 339.360016, 358.890015, 378.599976, 398.490021, 418.559998, 438.809998, 459.239990, 479.849976, 177.600006, 190.619995, 203.759995, 217.020004, 230.399994, 243.899994, 257.519989, 271.260010, 292.799988, 307.260010, 321.839996, 336.539978, 351.360016, 366.299988, 381.359985, 396.540009, 365.279999, 386.970001, 408.839996, 430.889984, 453.120026, 475.529968, 498.119995, 520.890015, 373.920013, 396.329987, 418.919983, 441.690002, 464.640015, 487.769958, 511.079987, 534.570007, 382.559998, 405.690002, 429.000000, 452.489990, 476.160004, 500.010010, 524.039978, 548.250000, 391.200012, 415.049988, 439.080017, 463.290009, 487.679993, 512.250000, 537.000000, 561.930054, 399.839996, 424.409973, 449.160034, 474.089966, 499.200012, 524.489990, 549.959961, 575.609985, 408.479980, 433.770020, 459.239990, 484.889954, 510.720032, 536.729980, 562.919983, 589.290039, 417.119995, 443.130005, 469.319977, 495.690002, 522.239990, 548.969971, 575.880005, 602.969971, 425.760010, 452.489990, 479.399994, 506.489990, 533.760010, 561.209961, 588.839966, 616.650024, 216.000000, 233.819992, 251.760010, 269.820007, 288.000000, 306.299988, 324.719971, 343.260010, 369.600006, 388.859985, 408.239990, 427.739990, 447.360016, 467.100006, 486.959961, 506.940002, 451.679993, 480.570007, 509.639984, 538.890015, 568.320007, 597.929993, 627.719971, 657.690002, 460.320007, 489.929993, 519.719971, 549.690002, 579.840027, 610.170044, 640.680054, 671.369995, 468.960022, 499.289978, 529.799988, 560.489990, 591.359985, 622.409973, 653.640015, 685.049988, 477.599976, 508.650024, 539.880005, 571.289978, 602.880005, 634.650024, 666.599976, 698.729980, 486.239990, 518.010010, 549.960022, 582.089966, 614.400024, 646.890015, 679.559937, 712.410034, 494.879974, 527.369995, 560.039978, 592.890015, 625.920044, 659.130005, 692.520020, 726.089966, 503.519989, 536.729980, 570.119995, 603.689941, 637.440063, 671.369995, 705.480042, 739.770020, 512.160034, 546.089966, 580.199951, 614.489990, 648.960022, 683.609985, 718.440002, 753.449951, 254.400009, 277.020020, 299.760010, 322.619995, 345.600006, 368.700012, 391.919983, 415.260010, 446.399994, 470.459961, 494.640015, 518.940002, 543.360046, 567.900024, 592.559998, 617.340027, 538.080017, 574.170044, 610.440002, 646.890015, 683.520020, 720.329956, 757.320007, 794.489990, 546.719971, 583.530029, 620.520020, 657.690002, 695.040039, 732.570007, 770.279968, 808.169983, 555.359985, 592.889954, 630.599976, 668.489990, 706.559998, 744.809998, 783.239990, 821.849976, 564.000000, 602.250000, 640.679993, 679.289978, 718.080017, 757.050049, 796.199951, 835.530029, 572.640015, 611.609985, 650.760010, 690.089966, 729.600037, 769.289978, 809.160034, 849.210083, 581.279968, 620.970032, 660.839966, 700.889954, 741.119995, 781.529968, 822.119995, 862.890015, 589.919983, 630.330017, 670.919983, 711.690002, 752.640015, 793.770020, 835.079956, 876.570007, 598.559998, 639.690002, 681.000000, 722.490051, 764.160034, 806.010010, 848.039978, 890.250061, 292.799988, 320.220001, 347.760010, 375.419983, 403.200012, 431.100006, 459.119995, 487.260010, 523.199951, 552.059998, 581.040039, 610.139954, 639.360046, 668.699951, 698.159973, 727.739990, 624.479980, 667.770020, 711.239990, 754.890015, 798.719971, 842.729980, 886.919983, 931.290039, 633.119995, 677.130005, 721.319946, 765.690002, 810.239990, 854.969971, 899.880005, 944.969971, 641.760010, 686.489990, 731.400024, 776.489990, 821.760010, 867.209961, 912.839966, 958.650024, 650.400024, 695.849976, 741.479980, 787.290

    input.linspace(-10, 0.1);
    weights.linspace(-2, 0.1);

    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->printBuffer();

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

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

    delete results;
}

//////////////////////////////////////////////////////////////////////
TEST_F(ConvolutionTests2, depthwise_conv2d_9) {

    int bS=1, iH=10,iW=10,  iC=8,mC=1,  kH=3,kW=3,  sH=1,sW=1,  pH=0,pW=0,  dH=1,dW=1;
    int       oC=iC*mC;
    int       oH=10,oW=10;
    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, mC}, nd4j::DataType::FLOAT32);

    NDArray expOutput('c', {bS, oC, oH, oW}, {-103.360001, -131.440002, -130.000000, -128.559998, -127.120003, -125.680000, -124.240005, -122.799995, -121.360001, -66.720001,-76.199997, -81.239998, -80.160004, -79.080002, -78.000000, -76.919998, -75.840004, -74.760002, -73.680000, -29.400002, -66.599998, -70.440002, -69.360001, -68.279999,
        -67.199997, -66.120003, -65.040001, -63.959999, -62.879997, -24.599997, -57.000000, -59.639999, -58.560005, -57.479996, -56.399998, -55.320000, -54.240002, -53.159996, -52.080002, -19.799997, -47.400002, -48.840000, -47.760002, -46.680000, -45.599998, -44.520000, -43.440002, -42.360001, -41.279999, -15.000000, -37.799999, -38.040001,
        -36.959999, -35.879997, -34.799999, -33.720001, -32.639999, -31.560001, -30.479996, -10.199999, -28.200001, -27.240002, -26.160000, -25.080002, -24.000000, -22.919998,-21.840000, -20.759998, -19.679998, -5.400000, -18.599998, -16.439999, -15.360001, -14.280001, -13.200001, -12.120001, -11.040000, -9.960001, -8.880000, -0.600000,
        -9.000000, -5.639999, -4.560000, -3.480000, -2.400000, -1.320001, -0.240000, 0.840001, 1.920000, 4.200000, 0.160000, 3.920000, 3.920000, 3.920000, 3.920000, 3.920000,3.920001, 3.920000, 3.920000, 3.520000, 8.860001, 12.920000, 14.420000, 15.920000, 17.420000, 18.920000, 20.420000, 21.920000, 23.420000, 13.820000, 20.430000, 27.750000,
        28.919998, 30.090000, 31.260000, 32.430000, 33.600002, 34.770000, 35.939999, 19.709999, 30.630001, 39.450001, 40.619999, 41.790001, 42.960003, 44.129997, 45.299999, 46.470001, 47.639999, 25.110001, 40.829998, 51.150002, 52.320000, 53.489998, 54.660004, 55.829994, 57.000000, 58.169998, 59.340004, 30.510002, 51.029999, 62.849998,
        64.019997, 65.190002, 66.360001, 67.529999, 68.699997, 69.870003, 71.040001, 35.910000, 61.229996, 74.550003, 75.720001, 76.889999, 78.059998, 79.229996, 80.400002, 81.570000, 82.740005, 41.310001, 71.430000, 86.250000, 87.419998, 88.589996, 89.760002, 90.929993, 92.099991, 93.270004, 94.440002, 46.709999, 81.630005, 97.949997,
        99.120003, 100.290009, 101.459999, 102.630005, 103.800003, 104.970001, 106.139999, 52.110001, 91.830002, 109.649994, 110.820007, 111.990005, 113.159996, 114.330002, 115.500000, 116.669998, 117.839996, 57.509995, 19.580000, 9.079998, 9.139999, 9.199999, 9.259996, 9.320001, 9.379998, 9.440000, 9.500000, -8.740000, 129.080002, 169.279999,
        170.839996, 172.399994, 173.960007, 175.520004, 177.080002, 178.639999, 180.199982, 102.360001, 129.059998, 154.739990, 156.000000, 157.259995, 158.520004, 159.779999, 161.039993, 162.300003, 163.559998, 80.820000, 139.860001, 167.340012, 168.600006, 169.860001, 171.119995, 172.380005, 173.639999, 174.899994, 176.160004, 86.820000,
        150.660004, 179.940002, 181.200012, 182.459991, 183.720001, 184.980011, 186.239990, 187.500000, 188.759995, 92.820007, 161.459991, 192.540009, 193.799988, 195.059998, 196.319992, 197.579987, 198.839996, 200.100006, 201.360001, 98.820000, 172.259995, 205.139999, 206.399994, 207.660004, 208.919983, 210.179993, 211.440002, 212.700012,
        213.959991, 104.819992, 183.059998, 217.739990, 219.000000, 220.259995, 221.519989, 222.779999, 224.039993, 225.300018, 226.559998, 110.819992, 193.860016, 230.339996, 231.600006, 232.860001, 234.119995, 235.380005, 236.639999, 237.900009, 239.160004, 116.820000, 204.660004, 242.940002, 244.199982, 245.459991, 246.720001, 247.980011,
        249.239990, 250.500000, 251.759995, 122.819992, 47.000000, 26.240004, 26.360004, 26.479998, 26.600002, 26.720001, 26.840002, 26.959997, 27.080000, -12.999998, 257.299988, 337.640015, 339.260010, 340.879974, 342.499969, 344.119995, 345.740021, 347.359985, 348.979980, 198.899994, 249.690002, 299.729980, 301.079987, 302.429993, 303.779999, 305.130005, 306.480011, 307.829987, 309.179993, 153.929993, 261.089996, 313.230011, 314.580017, 315.929993, 317.279968, 318.630005, 319.979980, 321.329987, 322.679993, 160.529999, 272.489990, 326.729980, 328.079987, 329.429993, 330.779968, 332.130005, 333.479980, 334.829987, 336.179993, 167.130005, 283.889984, 340.230011, 341.580017, 342.929993, 344.279999, 345.630005, 346.980011, 348.330017, 349.679993, 173.729996, 295.289978, 353.729980, 355.079987, 356.429993, 357.779968, 359.130005, 360.480011, 361.829987, 363.179993, 180.329987, 306.690002, 367.230011, 368.580017, 369.929993, 371.279999, 372.630005, 373.980011, 375.330017, 376.679993, 186.929993, 318.089996, 380.729980, 382.080017, 383.429993, 384.779968, 386.130005, 387.479980, 388.829987, 390.179993, 193.529984, 329.489990, 394.229980, 395.579987, 396.929993, 398.279999, 399.630005, 400.980011, 402.330017, 403.679993, 200.130005, 82.419998, 55.400005, 55.580002, 55.759995, 55.939999, 56.120003, 56.299995, 56.479996, 56.659996, -9.260002, 393.520020, 518.000000, 519.679993, 521.359985, 523.040039, 524.720032, 526.400024, 528.080017, 529.760010, 303.440002, 382.320007, 462.720032, 464.160004, 465.600037, 467.040009, 468.479980, 469.919983, 471.359985, 472.800018, 239.040009, 394.320007, 477.119995, 478.559998, 480.000000, 481.440002, 482.880005, 484.320007, 485.760010, 487.200012, 246.240005, 406.320007, 491.520020, 492.960022, 494.400024, 495.839996, 497.280029, 498.720032, 500.160004, 501.600037, 253.440002, 418.320007, 505.919983, 507.359985, 508.800018, 510.240051, 511.680023, 513.119995, 514.559998, 516.000000, 260.640015, 430.319977, 520.320007, 521.760010, 523.200012, 524.640015, 526.079956, 527.520020, 528.960022, 530.400024, 267.839996, 442.320007, 534.720032, 536.160034, 537.600037, 539.040039, 540.479980, 541.919983, 543.359985, 544.800049, 275.040009, 454.320007, 549.119995, 550.559998, 552.000000, 553.440002, 554.880005, 556.320007, 557.760010, 559.200012, 282.239990, 466.320007, 563.520020, 564.960022, 566.400024, 567.839966, 569.280029, 570.720032, 572.160034, 573.600037, 289.440002, 125.839996, 96.559998, 96.799995, 97.040009, 97.280014, 97.520004, 97.759995, 98.000000, 98.240013, 2.480007, 537.739990, 710.359985, 712.099976, 713.840027, 715.579956, 717.319946, 719.059998, 720.799988, 722.539978, 415.980011, 526.950012, 643.710022, 645.240051, 646.770020, 648.300049, 649.829956, 651.359985, 652.890015, 654.419983, 336.149994, 539.549988, 659.010010, 660.539978, 662.070007, 663.600037, 665.130005, 666.660034, 668.190002, 669.720032, 343.950012, 552.150024, 674.309998, 675.839966, 677.369995, 678.900024, 680.429993, 681.960022, 683.490051, 685.020020, 351.750000, 564.750000, 689.609985, 691.140015, 692.669983, 694.200012, 695.729980, 697.260010, 698.789978, 700.320007, 359.549988, 577.349976, 704.910034, 706.440002, 707.970032, 709.500000, 711.029968, 712.559998, 714.089966, 715.619995, 367.350037, 589.950012, 720.210022, 721.740051, 723.270020, 724.800049, 726.329956, 727.859985, 729.390015, 730.919983, 375.149994, 602.549988, 735.510010, 737.039978, 738.570007, 740.100037, 741.630005, 743.160034, 744.690002, 746.220032, 382.950012, 615.150024, 750.809998, 752.339966, 753.869995, 755.399963, 756.929993, 758.460022, 759.990051, 761.520020, 390.750000, 177.260010, 149.720001, 150.020004, 150.319992, 150.619995, 150.919998, 151.220001, 151.520004, 151.819992, 22.220009, 689.959961, 914.720032, 916.519958, 918.319946, 920.119995, 921.919983, 923.719971, 925.520020, 927.320007, 536.519958, 683.579956, 842.699951, 844.319946, 845.940002, 847.559998, 849.179993, 850.799988, 852.419983, 854.039978, 445.260010, 696.779968, 858.900024, 860.520020, 862.140015, 863.760010, 865.380005, 867.000000, 868.619995, 870.239990, 4

    input.linspace(-10, 0.1);
    weights.linspace(-2, 0.1);

    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);

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

    ASSERT_TRUE(expOutput.isSameShape(output));
    ASSERT_TRUE(expOutput.equalsTo(output, 1e-4));

    delete results;
}

#endif //LIBND4J_CONVOLUTIONTESTS2_H