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}, {2.75,   7.75,  12.75,  17.75,  22.75, 30.5 ,  40.5 ,  50.5 ,  60.5 ,  70.5 , 30.5 ,  40.5 ,  50.5 ,  60.5 ,  70.5 , 30.5 ,  40.5 ,  50.5 ,  60.5 ,  70.5 ,
                                                55.5 ,  65.5 ,  75.5 ,  85.5 ,  95.5 ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,
                                                55.5 ,  65.5 ,  75.5 ,  85.5 ,  95.5 ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,
                                                55.5 ,  65.5 ,  75.5 ,  85.5 ,  95.5 ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,
                                                 2.75,   7.75,  12.75,  17.75,  22.75, 30.5 ,  40.5 ,  50.5 ,  60.5 ,  70.5 , 30.5 ,  40.5 ,  50.5 ,  60.5 ,  70.5 , 30.5 ,  40.5 ,  50.5 ,  60.5 ,  70.5 ,
                                                55.5 ,  65.5 ,  75.5 ,  85.5 ,  95.5 ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,
                                                55.5 ,  65.5 ,  75.5 ,  85.5 ,  95.5 ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,
                                                55.5 ,  65.5 ,  75.5 ,  85.5 ,  95.5 ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  ,161.  , 181.  , 201.  , 221.  , 241.  });
    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, 8, 8, 16});

    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,   58827.4200139};
    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, 4806.747};
    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);
    weightsD.applyScalar(scalar::Divide, 100.0);
    weightsP.applyScalar(scalar::Divide, 100.0);
    epsilonNext.applyScalar(scalar::Divide, 100.0);

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

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

    const OpArgsHolder argsHolderFF({&input, &weights, &bias},         {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    const OpArgsHolder argsHolderBP({&input, &weights, &bias, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});

    nd4j::ops::deconv3d opFF;
    nd4j::ops::deconv3d_bp opBP;

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

    ASSERT_TRUE(isGradCorrect);
}

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

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

    const OpArgsHolder argsHolderFF({&input, &weights},         {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    const OpArgsHolder argsHolderBP({&input, &weights, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});

    nd4j::ops::deconv3d opFF;
    nd4j::ops::deconv3d_bp opBP;

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

    ASSERT_TRUE(isGradCorrect);
}

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

    input = 0.5;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.5);
    weights.permutei({2, 3, 4, 1, 0});

    const OpArgsHolder argsHolderFF({&input, &weights},         {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    const OpArgsHolder argsHolderBP({&input, &weights, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});

    nd4j::ops::deconv3d opFF;
    nd4j::ops::deconv3d_bp opBP;

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

    ASSERT_TRUE(isGradCorrect);
}

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

    input = 0.5;
    weights.linspace(0.1, 0.1);
    gradO.linspace(0.5);
    weights.permutei({2, 3, 4, 1, 0});

    const OpArgsHolder argsHolderFF({&input, &weights},         {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});
    const OpArgsHolder argsHolderBP({&input, &weights, &gradO}, {}, {kD,kH,kW,  sD,sH,sW,  pD,pH,pW,  dD,dH,dW, paddingMode, dataFormat});

    nd4j::ops::deconv3d opFF;
    nd4j::ops::deconv3d_bp opBP;

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

    ASSERT_TRUE(isGradCorrect);
}

//////////////////////////////////////////////////////////////////////
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}, {0.27620894, 0.21801452, 0.062078513, 7.348895E-4, 0.24149609, 0.4948205, 0.93483436, 0.52035654, 0.30292067, 0.3289706, 0.7977864,
                                                     0.03180518, 0.1455722, 0.90352905, 0.9405744, 0.0048329555, 0.44062102, 0.111197524, 0.31742015, 0.1933705, 0.23825112, 0.35076278, 0.7135856, 0.28229436, 0.18310733,
                                                     0.9613717, 0.56823575, 0.78289545, 0.62195826, 0.5244586, 0.5040889, 0.025349546, 0.41400263, 0.28420195, 0.8536445, 0.3044107, 0.7997134, 0.45762005, 0.7653578,
                                                     0.07198584, 0.5304998, 0.7334402, 0.85019743, 0.031957153, 0.37088063, 0.85722464, 0.06376881, 0.39791203});

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

    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}, {10.5, 11.5, 13.5, 14.5, 22.5, 23.5, 25.5, 26.5, 46.5, 47.5, 49.5, 50.5, 58.5, 59.5, 61.5, 62.5,
                                                         82.5, 83.5, 85.5, 86.5, 94.5, 95.5, 97.5, 98.5,118.5,119.5,121.5,122.5,130.5,131.5,133.5,134.5,
                                                        154.5,155.5,157.5,158.5,166.5,167.5,169.5,170.5,190.5,191.5,193.5,194.5,202.5,203.5,205.5,206.5});
    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. , 26. , 27. , 28. , 29. , 30. , 29.5, 30.5, 31.5, 29.5, 30.5, 31.5, 32.5, 33.5, 34.5, 34. , 35. , 36. , 38.5, 39.5, 40.5, 41.5, 42.5, 43.5, 43. , 44. , 45. , 43. , 44. , 45. , 46. , 47. , 48. , 47.5, 48.5, 49.5,
                                                           61. , 62. , 63. , 64. , 65. , 66. , 65.5, 66.5, 67.5, 65.5, 66.5, 67.5, 68.5, 69.5, 70.5, 70. , 71. , 72. , 74.5, 75.5, 76.5, 77.5, 78.5, 79.5, 79. , 80. , 81. , 79. , 80. , 81. , 82. , 83. , 84. , 83.5, 84.5, 85.5,
                                                           79. , 80. , 81. , 82. , 83. , 84. , 83.5, 84.5, 85.5, 83.5, 84.5, 85.5, 86.5, 87.5, 88.5, 88. , 89. , 90. , 92.5, 93.5, 94.5, 95.5, 96.5, 97.5, 97. , 98. , 99. , 97. , 98. , 99. ,100. ,101. ,102. ,101.5,102.5,103.5,
                                                          133. ,134. ,135. ,136. ,137. ,138. ,137.5,138.5,139.5,137.5,138.5,139.5,140.5,141.5,142.5,142. ,143. ,144. ,146.5,147.5,148.5,149.5,150.5,151.5,151. ,152. ,153. ,151. ,152. ,153. ,154. ,155. ,156. ,155.5,156.5,157.5,
                                                          169. ,170. ,171. ,172. ,173. ,174. ,173.5,174.5,175.5,173.5,174.5,175.5,176.5,177.5,178.5,178. ,179. ,180. ,182.5,183.5,184.5,185.5,186.5,187.5,187. ,188. ,189. ,187. ,188. ,189. ,190. ,191. ,192. ,191.5,192.5,193.5,
                                                          187. ,188. ,189. ,190. ,191. ,192. ,191.5,192.5,193.5,191.5,192.5,193.5,194.5,195.5,196.5,196. ,197. ,198. ,200.5,201.5,202.5,203.5,204.5,205.5,205. ,206. ,207. ,205. ,206. ,207. ,208. ,209. ,210. ,209.5,210.5,211.5});
    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}, {  29.5, 30.5, 31.5, 32.5, 33.5, 34.5, 38.5, 39.5, 40.5, 41.5, 42.5, 43.5, 65.5, 66.5, 67.5, 68.5, 69.5, 70.5,
                                                           74.5, 75.5, 76.5, 77.5, 78.5, 79.5,137.5,138.5,139.5,140.5,141.5,142.5,146.5,147.5,148.5,149.5,150.5,151.5,
                                                          173.5,174.5,175.5,176.5,177.5,178.5,182.5,183.5,184.5,185.5,186.5,187.5});
    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},{0.416667, 1.00, 1.333333, 0.75, 1.00, 2.25, 2.75, 1.50, 1.75, 3.75, 4.25, 2.25, 1.416667, 3.00, 3.333333, 1.75, 2.833333, 6.00, 6.666667, 3.50, 5.00, 10.50, 11.50, 6.00, 6.50,
                                                        13.50, 14.50, 7.50, 4.833333, 10.00, 10.666667, 5.50, 6.833333, 14.00, 14.666667, 7.50, 11.00, 22.50, 23.50, 12.00, 12.50, 25.50, 26.50, 13.50, 8.833333, 18.00, 18.666666, 9.50,
                                                        4.416667, 9.00, 9.333333, 4.75, 7.00, 14.25, 14.75, 7.50, 7.75, 15.75, 16.25, 8.25, 5.416667, 11.00, 11.333333, 5.75, 6.416667, 13.00, 13.333333, 6.75, 10.00, 20.25, 20.75,
                                                        10.50, 10.75, 21.75, 22.25, 11.25, 7.416667, 15.00, 15.333333, 7.75, 14.833333, 30.00, 30.666666, 15.50, 23.00, 46.50, 47.50, 24.00, 24.50, 49.50, 50.50, 25.50, 16.833334,
                                                        34.00, 34.666668, 17.50, 18.833334, 38.00, 38.666668, 19.50, 29.00, 58.50, 59.50, 30.00, 30.50, 61.50, 62.50, 31.50, 20.833334, 42.00, 42.666668, 21.50, 10.416667, 21.00,
                                                        21.333334, 10.75, 16.00, 32.25, 32.75, 16.50, 16.75, 33.75, 34.25, 17.25, 11.416667, 23.00, 23.333334, 11.75, 12.416667, 25.00, 25.333334, 12.75, 19.00, 38.25, 38.75, 19.50,
                                                        19.75, 39.75, 40.25, 20.25, 13.416667, 27.00, 27.333334, 13.75, 26.833334, 54.00, 54.666668, 27.50, 41.00, 82.50, 83.50, 42.00, 42.50, 85.50, 86.50, 43.50, 28.833334, 58.00,
                                                        58.666668, 29.50, 30.833334, 62.00, 62.666668, 31.50, 47.00, 94.50, 95.50, 48.00, 48.50, 97.50, 98.50, 49.50, 32.833332, 66.00, 66.666664, 33.50, 16.416666, 33.00, 33.333332,
                                                        16.75, 25.00, 50.25, 50.75, 25.50, 25.75, 51.75, 52.25, 26.25, 17.416666, 35.00, 35.333332, 17.75, 18.416666, 37.00, 37.333332, 18.75, 28.00, 56.25, 56.75, 28.50, 28.75,
                                                        57.75, 58.25, 29.25, 19.416666, 39.00, 39.333332, 19.75, 38.833332, 78.00, 78.666664, 39.50, 59.00, 118.50, 119.50, 60.00, 60.50, 121.50, 122.50, 61.50, 40.833332, 82.00,
                                                        82.666664, 41.50, 42.833332, 86.00, 86.666664, 43.50, 65.00, 130.50, 131.50, 66.00, 66.50, 133.50, 134.50, 67.50, 44.833332, 90.00, 90.666664, 45.50, 22.416666, 45.00,
                                                        45.333332, 22.75, 34.00, 68.25, 68.75, 34.50, 34.75, 69.75, 70.25, 35.25, 23.416666, 47.00, 47.333332, 23.75, 24.416666, 49.00, 49.333332, 24.75, 37.00, 74.25, 74.75,
                                                        37.50, 37.75, 75.75, 76.25, 38.25, 25.416666, 51.00, 51.333332, 25.75, 50.833332, 102.00, 102.666664, 51.50, 77.00, 154.50, 155.50, 78.00, 78.50, 157.50, 158.50, 79.50,
                                                        52.833332, 106.00, 106.666664, 53.50, 54.833332, 110.00, 110.666664, 55.50, 83.00, 166.50, 167.50, 84.00, 84.50, 169.50, 170.50, 85.50, 56.833332, 114.00, 114.666664,
                                                        57.50, 28.416666, 57.00, 57.333332, 28.75, 43.00, 86.25, 86.75, 43.50, 43.75, 87.75, 88.25, 44.25, 29.416666, 59.00, 59.333332, 29.75, 30.416666, 61.00, 61.333332, 30.75,
                                                        46.00, 92.25, 92.75, 46.50, 46.75, 93.75, 94.25, 47.25, 31.416666, 63.00, 63.333332, 31.75, 62.833332, 126.00, 126.666664, 63.50, 95.00, 190.50, 191.50, 96.00, 96.50,
                                                        193.50, 194.50, 97.50, 64.833336, 130.00, 130.666672, 65.50, 66.833336, 134.00, 134.666672, 67.50, 101.00, 202.50, 203.50, 102.00, 102.50, 205.50, 206.50, 103.50,
                                                        68.833336, 138.00, 138.666672, 69.50, 34.416668, 69.00, 69.333336, 34.75, 52.00, 104.25, 104.75, 52.50, 52.75, 105.75, 106.25, 53.25, 35.416668, 71.00, 71.333336, 35.75});
    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., 21., 23., 24., 32., 33., 35., 36., 56., 57., 59., 60., 68., 69., 71., 72., 92., 93., 95., 96.,104.,105.,107.,108.,
                                                         128.,129.,131.,132.,140.,141.,143.,144.,164.,165.,167.,168.,176.,177.,179.,180.,200.,201.,203.,204.,212.,213.,215.,216.});
    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.,  50.,  51., 52.,  53.,  54., 52.,  53.,  54., 58.,  59.,  60., 61.,  62.,  63., 61.,  62.,  63., 67.,  68.,  69., 70.,  71.,  72., 70.,  71.,  72., 67.,  68.,  69., 70.,  71.,  72., 70.,  71.,  72.,
                                                          85.,  86.,  87., 88.,  89.,  90., 88.,  89.,  90., 94.,  95.,  96., 97.,  98.,  99., 97.,  98.,  99.,103., 104., 105.,106., 107., 108.,106., 107., 108.,103., 104., 105.,106., 107., 108.,106., 107., 108.,
                                                          85.,  86.,  87., 88.,  89.,  90., 88.,  89.,  90., 94.,  95.,  96., 97.,  98.,  99., 97.,  98.,  99.,103., 104., 105.,106., 107., 108.,106., 107., 108.,103., 104., 105.,106., 107., 108.,106., 107., 108.,
                                                         157., 158., 159.,160., 161., 162.,160., 161., 162.,166., 167., 168.,169., 170., 171.,169., 170., 171.,175., 176., 177.,178., 179., 180.,178., 179., 180.,175., 176., 177.,178., 179., 180.,178., 179., 180.,
                                                         193., 194., 195.,196., 197., 198.,196., 197., 198.,202., 203., 204.,205., 206., 207.,205., 206., 207.,211., 212., 213.,214., 215., 216.,214., 215., 216.,211., 212., 213.,214., 215., 216.,214., 215., 216.,
                                                         193., 194., 195.,196., 197., 198.,196., 197., 198.,202., 203., 204.,205., 206., 207.,205., 206., 207.,211., 212., 213.,214., 215., 216.,214., 215., 216.,211., 212., 213.,214., 215., 216.,214., 215., 216.});
    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.,  59.,  60., 61.,  62.,  63., 67.,  68.,  69., 70.,  71.,  72., 94.,  95.,  96., 97.,  98.,  99.,103., 104., 105.,106., 107., 108.,
                                                         166., 167., 168.,169., 170., 171.,175., 176., 177.,178., 179., 180.,202., 203., 204.,205., 206., 207.,211., 212., 213.,214., 215., 216.});
    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.,   5.,   6.,   6.,  7.,   8.,   9.,   9., 10.,  11.,  12.,  12., 10.,  11.,  12.,  12., 16.,  17.,  18.,  18., 19.,  20.,  21.,  21., 22.,  23.,  24.,  24., 22.,  23.,  24.,  24., 28.,  29.,  30.,  30., 31.,  32.,  33.,  33., 34.,  35.,  36.,  36., 34.,  35.,  36.,  36.,
                                                         28.,  29.,  30.,  30., 31.,  32.,  33.,  33., 34.,  35.,  36.,  36., 34.,  35.,  36.,  36., 40.,  41.,  42.,  42., 43.,  44.,  45.,  45., 46.,  47.,  48.,  48., 46.,  47.,  48.,  48., 52.,  53.,  54.,  54., 55.,  56.,  57.,  57., 58.,  59.,  60.,  60., 58.,  59.,  60.,  60.,
                                                         64.,  65.,  66.,  66., 67.,  68.,  69.,  69., 70.,  71.,  72.,  72., 70.,  71.,  72.,  72., 64.,  65.,  66.,  66., 67.,  68.,  69.,  69., 70.,  71.,  72.,  72., 70.,  71.,  72.,  72., 76.,  77.,  78.,  78., 79.,  80.,  81.,  81., 82.,  83.,  84.,  84., 82.,  83.,  84.,  84.,
                                                         88.,  89.,  90.,  90., 91.,  92.,  93.,  93., 94.,  95.,  96.,  96., 94.,  95.,  96.,  96.,100., 101., 102., 102.,103., 104., 105., 105.,106., 107., 108., 108.,106., 107., 108., 108.,100., 101., 102., 102.,103., 104., 105., 105.,106., 107., 108., 108.,106., 107., 108., 108.,
                                                        112., 113., 114., 114.,115., 116., 117., 117.,118., 119., 120., 120.,118., 119., 120., 120.,124., 125., 126., 126.,127., 128., 129., 129.,130., 131., 132., 132.,130., 131., 132., 132.,136., 137., 138., 138.,139., 140., 141., 141.,142., 143., 144., 144.,142., 143., 144., 144.,
                                                        136., 137., 138., 138.,139., 140., 141., 141.,142., 143., 144., 144.,142., 143., 144., 144.,148., 149., 150., 150.,151., 152., 153., 153.,154., 155., 156., 156.,154., 155., 156., 156.,160., 161., 162., 162.,163., 164., 165., 165.,166., 167., 168., 168.,166., 167., 168., 168.,
                                                        172., 173., 174., 174.,175., 176., 177., 177.,178., 179., 180., 180.,178., 179., 180., 180.,172., 173., 174., 174.,175., 176., 177., 177.,178., 179., 180., 180.,178., 179., 180., 180.,184., 185., 186., 186.,187., 188., 189., 189.,190., 191., 192., 192.,190., 191., 192., 192.,
                                                        196., 197., 198., 198.,199., 200., 201., 201.,202., 203., 204., 204.,202., 203., 204., 204.,208., 209., 210., 210.,211., 212., 213., 213.,214., 215., 216., 216.,214., 215., 216., 216.,208., 209., 210., 210.,211., 212., 213., 213.,214., 215., 216., 216.,214., 215., 216., 216.});
    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}, {0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.666667, 1.333333, 0.666667,0.666667, 1.333333, 0.666667,0.333333, 0.666667, 0.333333,
                                                         0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,
                                                         0.333333, 0.666667, 0.333333,0.666667, 1.333333, 0.666667,0.666667, 1.333333, 0.666667,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,
                                                         0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.666667, 1.333333, 0.666667,0.666667, 1.333333, 0.666667,0.333333, 0.666667, 0.333333,
                                                         0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,
                                                         0.333333, 0.666667, 0.333333,0.666667, 1.333333, 0.666667,0.666667, 1.333333, 0.666667,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,
                                                         0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.666667, 1.333333, 0.666667,0.666667, 1.333333, 0.666667,0.333333, 0.666667, 0.333333,
                                                         0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,
                                                         0.333333, 0.666667, 0.333333,0.666667, 1.333333, 0.666667,0.666667, 1.333333, 0.666667,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667,0.333333, 0.666667, 0.333333,0.333333, 0.666667, 0.333333,0.166667, 0.333333, 0.166667});
    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}, {1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,
                                                         1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,
                                                         1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,
                                                         1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,
                                                         1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,
                                                         1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,
                                                         1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,
                                                         1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,
                                                         1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333,1.333333, 1.333333, 1.333333,2. , 2. , 2. ,2. , 2. , 2. ,1.333333, 1.333333, 1.333333});
    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}, {0.41667, 0.41667, 0.41667,0.83333, 0.83333, 0.83333,1.25, 1.25, 1.25 ,0.58333, 0.58333, 0.58333,1.16667, 1.16667, 1.16667,1.75, 1.75, 1.75 ,0.58333, 0.58333, 0.58333,1.16667, 1.16667, 1.16667,1.75, 1.75, 1.75 ,
                                                         0.41667, 0.41667, 0.41667,0.83333, 0.83333, 0.83333,1.25, 1.25, 1.25 ,0.83333, 0.83333, 0.83333,1.66667, 1.66667, 1.66667,2.5 , 2.5 , 2.5  ,1.16667, 1.16667, 1.16667,2.33333, 2.33333, 2.33333,3.5 , 3.5 , 3.5  ,
                                                         1.16667, 1.16667, 1.16667,2.33333, 2.33333, 2.33333,3.5 , 3.5 , 3.5  ,0.83333, 0.83333, 0.83333,1.66667, 1.66667, 1.66667,2.5 , 2.5 , 2.5  ,1.25   , 1.25   , 1.25   ,2.5    , 2.5    , 2.5    ,3.75, 3.75, 3.75 ,
                                                         1.75   , 1.75   , 1.75   ,3.5    , 3.5    , 3.5    ,5.25, 5.25, 5.25 ,1.75   , 1.75   , 1.75   ,3.5    , 3.5    , 3.5    ,5.25, 5.25, 5.25 ,1.25   , 1.25   , 1.25   ,2.5    , 2.5    , 2.5    ,3.75, 3.75, 3.75 ,
                                                         0.41667, 0.41667, 0.41667,0.83333, 0.83333, 0.83333,1.25, 1.25, 1.25 ,0.58333, 0.58333, 0.58333,1.16667, 1.16667, 1.16667,1.75, 1.75, 1.75 ,0.58333, 0.58333, 0.58333,1.16667, 1.16667, 1.16667,1.75, 1.75, 1.75 ,
                                                         0.41667, 0.41667, 0.41667,0.83333, 0.83333, 0.83333,1.25, 1.25, 1.25 ,0.83333, 0.83333, 0.83333,1.66667, 1.66667, 1.66667,2.5 , 2.5 , 2.5  ,1.16667, 1.16667, 1.16667,2.33333, 2.33333, 2.33333,3.5 , 3.5 , 3.5  ,
                                                         1.16667, 1.16667, 1.16667,2.33333, 2.33333, 2.33333,3.5 , 3.5 , 3.5  ,0.83333, 0.83333, 0.83333,1.66667, 1.66667, 1.66667,2.5 , 2.5 , 2.5  ,1.25   , 1.25   , 1.25   ,2.5    , 2.5    , 2.5    ,3.75, 3.75, 3.75 ,
                                                         1.75   , 1.75   , 1.75   ,3.5    , 3.5    , 3.5    ,5.25, 5.25, 5.25 ,1.75   , 1.75   , 1.75   ,3.5    , 3.5    , 3.5    ,5.25, 5.25, 5.25 ,1.25   , 1.25   , 1.25   ,2.5    , 2.5    , 2.5    ,3.75, 3.75, 3.75 });
    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}, {0.16667, 0.16667, 0.16667,0.33333, 0.33333, 0.33333,0.5 , 0.5  , 0.5  ,0.33333, 0.33333, 0.33333,0.66667, 0.66667, 0.66667,1.  , 1.   , 1.   ,0.58333, 0.58333, 0.58333,1.16667, 1.16667, 1.16667,1.75, 1.75 , 1.75 ,
                                                         0.91667, 0.91667, 0.91667,1.83333, 1.83333, 1.83333,2.75, 2.75 , 2.75 ,0.33333, 0.33333, 0.33333,0.66667, 0.66667, 0.66667,1.  , 1.   , 1.   ,0.66667, 0.66667, 0.66667,1.33333, 1.33333, 1.33333,2.  , 2.   , 2.   ,
                                                         1.16667, 1.16667, 1.16667,2.33333, 2.33333, 2.33333,3.5 , 3.5  , 3.5  ,1.83333, 1.83333, 1.83333,3.66667, 3.66667, 3.66667,5.5 , 5.5  , 5.5  ,0.5    , 0.5    , 0.5    ,1.     , 1.     , 1.     ,1.5 , 1.5  , 1.5  ,
                                                         1.     , 1.     , 1.     ,2.     , 2.     , 2.     ,3.  , 3.   , 3.   ,1.75   , 1.75   , 1.75   ,3.5    , 3.5    , 3.5    ,5.25, 5.25 , 5.25 ,2.75   , 2.75   , 2.75   ,5.5    , 5.5    , 5.5    ,8.25, 8.25 , 8.25 ,
                                                         0.16667, 0.16667, 0.16667,0.33333, 0.33333, 0.33333,0.5 , 0.5  , 0.5  ,0.33333, 0.33333, 0.33333,0.66667, 0.66667, 0.66667,1.  , 1.   , 1.   ,0.58333, 0.58333, 0.58333,1.16667, 1.16667, 1.16667,1.75, 1.75 , 1.75 ,
                                                         0.91667, 0.91667, 0.91667,1.83333, 1.83333, 1.83333,2.75, 2.75 , 2.75 ,0.33333, 0.33333, 0.33333,0.66667, 0.66667, 0.66667,1.  , 1.   , 1.   ,0.66667, 0.66667, 0.66667,1.33333, 1.33333, 1.33333,2.  , 2.   , 2.   ,
                                                         1.16667, 1.16667, 1.16667,2.33333, 2.33333, 2.33333,3.5 , 3.5  , 3.5  ,1.83333, 1.83333, 1.83333,3.66667, 3.66667, 3.66667,5.5 , 5.5  , 5.5  ,0.5    , 0.5    , 0.5    ,1.     , 1.     , 1.     ,1.5 , 1.5  , 1.5  ,
                                                         1.     , 1.     , 1.     ,2.     , 2.     , 2.     ,3.  , 3.   , 3.   ,1.75   , 1.75   , 1.75   ,3.5    , 3.5    , 3.5    ,5.25, 5.25 , 5.25 ,2.75   , 2.75   , 2.75   ,5.5    , 5.5    , 5.5    ,8.25, 8.25 , 8.25 });
    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. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0.1, 0.2,0. , 0.3, 0.4,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0.5, 0.6,0. , 0.7, 0.8,
                                                         0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0.9, 1. ,0. , 1.1, 1.2,0. , 0. , 0. ,0. , 0. , 0. ,0. , 1.3, 1.4,0. , 1.5, 1.6,
                                                         0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 1.7, 1.8,0. , 1.9, 2. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 2.1, 2.2,0. , 2.3, 2.4,
                                                         0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 2.5, 2.6,0. , 2.7, 2.8,0. , 0. , 0. ,0. , 0. , 0. ,0. , 2.9, 3. ,0. , 3.1, 3.2,
                                                         0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 3.3, 3.4,0. , 3.5, 3.6,0. , 0. , 0. ,0. , 0. , 0. ,0. , 3.7, 3.8,0. , 3.9, 4. ,
                                                         0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 4.1, 4.2,0. , 4.3, 4.4,0. , 0. , 0. ,0. , 0. , 0. ,0. , 4.5, 4.6,0. , 4.7, 4.8});
    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+00, 0.000e+00, 0.000e+00,1.000e-01, 2.000e-01, 7.000e-01,5.000e-01, 6.000e-01, 1.500e+00,2.200e+00, 2.400e+00, 5.400e+00,0.000e+00, 0.000e+00, 0.000e+00,1.700e+00, 1.800e+00, 3.900e+00,2.100e+00, 2.200e+00, 4.700e+00,5.400e+00, 5.600e+00, 1.180e+01,
                                                         0.000e+00, 0.000e+00, 0.000e+00,8.200e+00, 8.400e+00, 1.740e+01,9.000e+00, 9.200e+00, 1.900e+01,2.040e+01, 2.080e+01, 4.280e+01,0.000e+00, 0.000e+00, 0.000e+00,6.500e+00, 6.600e+00, 1.350e+01,6.900e+00, 7.000e+00, 1.430e+01,1.500e+01, 1.520e+01, 3.100e+01,
                                                         0.000e+00, 0.000e+00, 0.000e+00,8.100e+00, 8.200e+00, 1.670e+01,8.500e+00, 8.600e+00, 1.750e+01,1.820e+01, 1.840e+01, 3.740e+01,0.000e+00, 0.000e+00, 0.000e+00,2.100e+01, 2.120e+01, 4.300e+01,2.180e+01, 2.200e+01, 4.460e+01,4.600e+01, 4.640e+01, 9.400e+01,
                                                         0.000e+00, 0.000e+00, 0.000e+00,1.290e+01, 1.300e+01, 2.630e+01,1.330e+01, 1.340e+01, 2.710e+01,2.780e+01, 2.800e+01, 5.660e+01,0.000e+00, 0.000e+00, 0.000e+00,1.450e+01, 1.460e+01, 2.950e+01,1.490e+01, 1.500e+01, 3.030e+01,3.100e+01, 3.120e+01, 6.300e+01,
                                                         0.000e+00, 0.000e+00, 0.000e+00,3.380e+01, 3.400e+01, 6.860e+01,3.460e+01, 3.480e+01, 7.020e+01,7.160e+01, 7.200e+01, 1.452e+02,0.000e+00, 0.000e+00, 0.000e+00,1.930e+01, 1.940e+01, 3.910e+01,1.970e+01, 1.980e+01, 3.990e+01,4.060e+01, 4.080e+01, 8.220e+01,
                                                         0.000e+00, 0.000e+00, 0.000e+00,2.090e+01, 2.100e+01, 4.230e+01,2.130e+01, 2.140e+01, 4.310e+01,4.380e+01, 4.400e+01, 8.860e+01,0.000e+00, 0.000e+00, 0.000e+00,4.660e+01, 4.680e+01, 9.420e+01,4.740e+01, 4.760e+01, 9.580e+01,9.720e+01, 9.760e+01, 1.964e+02,
                                                         0.000e+00, 0.000e+00, 0.000e+00,2.570e+01, 2.580e+01, 5.190e+01,2.610e+01, 2.620e+01, 5.270e+01,5.340e+01, 5.360e+01, 1.078e+02,0.000e+00, 0.000e+00, 0.000e+00,2.730e+01, 2.740e+01, 5.510e+01,2.770e+01, 2.780e+01, 5.590e+01,5.660e+01, 5.680e+01, 1.142e+02,
                                                         0.000e+00, 0.000e+00, 0.000e+00,5.940e+01, 5.960e+01, 1.198e+02,6.020e+01, 6.040e+01, 1.214e+02,1.228e+02, 1.232e+02, 2.476e+02,0.000e+00, 0.000e+00, 0.000e+00,3.210e+01, 3.220e+01, 6.470e+01,3.250e+01, 3.260e+01, 6.550e+01,6.620e+01, 6.640e+01, 1.334e+02,
                                                         0.000e+00, 0.000e+00, 0.000e+00,3.370e+01, 3.380e+01, 6.790e+01,3.410e+01, 3.420e+01, 6.870e+01,6.940e+01, 6.960e+01, 1.398e+02,0.000e+00, 0.000e+00, 0.000e+00,7.220e+01, 7.240e+01, 1.454e+02,7.300e+01, 7.320e+01, 1.470e+02,1.484e+02, 1.488e+02, 2.988e+02});
    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., 0., 0.,  0. ,  0. ,  0. ,  0. ,   0.  ,  0. ,  0. , 0. ,  0. ,  0. ,   0.  ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,
                                                          0., 0., 0.,  0. ,  0. ,  0. ,  0. ,   0.  ,  0. ,  0. , 0. ,  0. ,  0. ,   0.  ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,  0.1,   0.2    ,  0.3,  1.1,   1.3    ,  1.5,
                                                          0., 0., 0.,  1. ,  1.1,  1.2,  2.9,   3.1 ,  3.3,  0. , 0. ,  0. ,  4.7,   4.9 ,  5.1, 11.2,  11.6    , 12. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,
                                                          0., 0., 0., 11. , 11.2, 11.4, 23.8,  24.2 , 24.6,  0. , 0. ,  0. , 12.8,  13.  , 13.2, 27.4,  27.8    , 28.2,  0. ,   0.     ,  0. , 31. ,  31.4    , 31.8, 65.6,  66.39999, 67.2,
                                                          0., 0., 0.,  0. ,  0. ,  0. ,  0. ,   0.  ,  0. ,  0. , 0. ,  0. ,  0. ,   0.  ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,
                                                          0., 0., 0.,  0. ,  0. ,  0. ,  0. ,   0.  ,  0. ,  0. , 0. ,  0. ,  0. ,   0.  ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. , 10.9,  11.     , 11.1, 22.7,  22.9    , 23.1,
                                                          0., 0., 0., 11.8, 11.9, 12. , 24.5,  24.7 , 24.9,  0. , 0. ,  0. , 26.3,  26.5 , 26.7, 54.4,  54.8    , 55.2,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,  0. ,   0.     ,  0. ,
                                                          0., 0., 0., 32.6, 32.8, 33. , 67. ,  67.4 , 67.8,  0. , 0. ,  0. , 34.4,  34.6 , 34.8, 70.6,  71.     , 71.4,  0. ,   0.     ,  0. , 74.2,  74.6    , 75. ,152. , 152.8    ,153.6});
    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, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                                                         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.1, 0.2, 0.3, 1.1, 1.3, 1.5, 0, 0, 0, 5.7, 6, 6.3,
                                                         14.1, 14.7, 15.3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 11, 11.2, 11.4, 23.8, 24.2,
                                                         24.6, 0, 0, 0, 43.8, 44.4, 45, 93, 94.2, 95.4, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                                                         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                                                         10.9, 11, 11.1, 22.7, 22.9, 23.1, 0, 0, 0, 38.1, 38.4, 38.7, 78.9, 79.5, 80.1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                                                         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 32.6, 32.8, 33, 67, 67.4, 67.8, 0, 0, 0, 108.6, 109.2, 109.8, 222.6, 223.8, 225,});
    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. , 0. , 0. ,0. , 0. , 0. ,0. , 0.1, 0.2,0. , 0.3, 0.4,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0.5, 0.6,0. , 0.7, 0.8,
                                                     0. , 0. , 0. ,0. , 0. , 0. ,0. , 0.9, 1. ,0. , 1.1, 1.2,0. , 0. , 0. ,0. , 0. , 0. ,0. , 1.3, 1.4,0. , 1.5, 1.6,
                                                     0. , 0. , 0. ,0. , 0. , 0. ,0. , 1.7, 1.8,0. , 1.9, 2. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 2.1, 2.2,0. , 2.3, 2.4});
    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. ,  0. ,  0. , 0.1,  0.2,  0.7, 0.5,  0.6,  1.5, 2.2,  2.4,  5.4, 0. ,  0. ,  0. , 1.7,  1.8,  3.9, 2.1,  2.2,  4.7, 5.4,  5.6, 11.8,
                                                     0. ,  0. ,  0. , 3.3,  3.4,  7.1, 3.7,  3.8,  7.9, 8.6,  8.8, 18.2, 0. ,  0. ,  0. , 4.9,  5. , 10.3, 5.3,  5.4, 11.1,11.8, 12. , 24.6,
                                                     0. ,  0. ,  0. , 6.5,  6.6, 13.5, 6.9,  7. , 14.3,15. , 15.2, 31. , 0. ,  0. ,  0. , 8.1,  8.2, 16.7, 8.5,  8.6, 17.5,18.2, 18.4, 37.4});
    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. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0.1, 0.2, 0.3, 1.1, 1.3, 1.5, 0. , 0. , 0. , 1. , 1.1, 1.2, 2.9, 3.1, 3.3,
                                                     0. , 0. , 0. , 4.7, 4.9, 5.1,11.2,11.6,12. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0. , 3.7, 3.8, 3.9, 8.3, 8.5, 8.7,
                                                     0. , 0. , 0. , 4.6, 4.7, 4.8,10.1,10.3,10.5, 0. , 0. , 0. ,11.9,12.1,12.3,25.6,26. ,26.4});
    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. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0.1, 0.2, 0.3,0.4, 0.5, 0.6,
                                                     0. , 0. , 0. ,0.7, 0.8, 0.9,1. , 1.1, 1.2,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,0. , 0. , 0. ,
                                                     0. , 0. , 0. ,1.3, 1.4, 1.5,1.6, 1.7, 1.8,0. , 0. , 0. ,1.9, 2. , 2.1,2.2, 2.3, 2.4});
    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}, {3.5 , 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});
    auto expected = NDArrayFactory::create<TypeParam>('c', {bS,iD,iH,iW}, {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});

    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}, {0.016667,0.05    ,0.033333,0.066667,0.166667,0.1     ,0.066667,0.166667,0.1     ,0.05    ,0.116667,0.066667,
                                                     0.083333,0.183333,0.1     ,0.2     ,0.433333,0.233333,0.2     ,0.433333,0.233333,0.116667,0.25    ,0.133333,
                                                     0.15    ,0.316667,0.166667,0.333333,0.7     ,0.366667,0.333333,0.7     ,0.366667,0.183333,0.383333,0.2     ,
                                                     0.216667,0.45    ,0.233333,0.466667,0.966667,0.5     ,0.466667,0.966667,0.5     ,0.25    ,0.516667,0.266667,
                                                     0.283333,0.583333,0.3     ,0.6     ,1.233333,0.633333,0.6     ,1.233333,0.633333,0.316667,0.65    ,0.333333,
                                                     0.35    ,0.716667,0.366667,0.733333,1.5     ,0.766667,0.733333,1.5     ,0.766667,0.383333,0.783333,0.4     });
    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}, {0.233333,0.3     ,0.366667,0.55    ,0.65    ,0.75    ,0.95    ,1.05    ,1.15    ,0.766667,0.833333,0.9     ,
                                                     1.3     ,1.366667,1.433333,2.15    ,2.25    ,2.35    ,2.55    ,2.65    ,2.75    ,1.833333,1.9     ,1.966667,
                                                     2.366667,2.433333,2.5     ,3.75    ,3.85    ,3.95    ,4.15    ,4.25    ,4.35    ,2.9     ,2.966667,3.033333,
                                                     3.433333,3.5     ,3.566667,5.35    ,5.45    ,5.55    ,5.75    ,5.85    ,5.95    ,3.966667,4.033333,4.1     ,
                                                     4.5     ,4.566667,4.633333,6.95    ,7.05    ,7.15    ,7.35    ,7.45    ,7.55    ,5.033333,5.1     ,5.166667,
                                                     5.566667,5.633333,5.7     ,8.549999,8.65    ,8.75    ,8.95    ,9.05    ,9.150001,6.1     ,6.166667,6.233334});
    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}, {0.19167, 0.23333, 0.275, 0.50833, 0.59167, 0.675, 1.2  , 1.325, 1.45 ,0.50833,0.56667, 0.625, 1.19167,1.30833, 1.425, 2.4  ,2.575, 2.75 ,
                                                     1.18333, 1.24167, 1.3  , 2.54167, 2.65833, 2.775, 4.425, 4.6  , 4.775,1.01667,1.05833, 1.1  , 2.15833,2.24167, 2.325, 3.675,3.8  , 3.925,
                                                     1.69167, 1.73333, 1.775, 3.50833, 3.59167, 3.675, 5.7  , 5.825, 5.95 ,2.60833,2.66667, 2.725, 5.39167,5.50833, 5.625, 8.7  ,8.875, 9.05 ,
                                                     3.28333, 3.34167, 3.4  , 6.74167, 6.85833, 6.975,10.725,10.9  ,11.075,2.51667,2.55833, 2.6  , 5.15833,5.24167, 5.325, 8.175,8.3  , 8.425});
    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}, {0.01667,0.03333,0.05,0.08333,0.11667,0.15,0.06667,0.08333,0.1,0.13333,0.16667,0.2 ,0.36667,0.43333,0.5 ,0.23333,0.26667,0.3,
                                                     0.13333,0.16667,0.2 ,0.36667,0.43333,0.5 ,0.23333,0.26667,0.3,0.11667,0.13333,0.15,0.28333,0.31667,0.35,0.16667,0.18333,0.2,
                                                     0.21667,0.23333,0.25,0.48333,0.51667,0.55,0.26667,0.28333,0.3,0.53333,0.56667,0.6 ,1.16667,1.23333,1.3 ,0.63333,0.66667,0.7,
                                                     0.53333,0.56667,0.6 ,1.16667,1.23333,1.3 ,0.63333,0.66667,0.7,0.31667,0.33333,0.35,0.68333,0.71667,0.75,0.36667,0.38333,0.4});
    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-04,9.671602e-03,1.306569e-02,3.679184e-02,1.297220e-01,1.040181e-01,1.126750e-01,3.320884e-01,2.340406e-01,1.333333e-01,3.352886e-01,2.070211e-01,
                                                     8.991618e-02,2.160601e-01,1.283173e-01,2.744226e-01,6.364498e-01,3.662123e-01,3.869788e-01,8.808994e-01,4.984556e-01,2.613189e-01,5.818475e-01,3.225517e-01,
                                                     2.065654e-01,4.553546e-01,2.501175e-01,5.190718e-01,1.131343e+00,6.148388e-01,6.362602e-01,1.377521e+00,7.439550e-01,3.833026e-01,8.227519e-01,4.407146e-01,
                                                     3.261206e-01,6.969233e-01,3.717564e-01,7.627507e-01,1.620991e+00,8.600952e-01,8.814538e-01,1.866888e+00,9.873542e-01,5.046682e-01,1.064004e+00,5.602558e-01,
                                                     4.464697e-01,9.389536e-01,4.932274e-01,1.005908e+00,2.108550e+00,1.104095e+00,1.125322e+00,2.354009e+00,1.230180e+00,6.258913e-01,1.305581e+00,6.804127e-01,
                                                     5.671396e-01,1.181128e+00,6.145977e-01,1.248783e+00,2.595083e+00,1.347494e+00,1.368600e+00,2.840157e+00,1.472778e+00,7.470673e-01,1.547362e+00,8.008900e-01});
    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}, {0.007931,0.042891,0.040544,0.09369 ,0.276841,0.191675,0.163957,0.442946,0.287512,0.154919,0.373153,0.221172,
                                                     0.15901 ,0.365232,0.207846,0.428282,0.959455,0.534076,0.508585,1.128771,0.623089,0.319794,0.698063,0.379547,
                                                     0.321068,0.692438,0.372316,0.757521,1.620323,0.864566,0.838684,1.787943,0.951023,0.483194,1.023434,0.541058,
                                                     0.483937,1.019414,0.536145,1.085348,2.276996,1.192917,1.166749,2.443606,1.278126,0.646499,1.349361,0.703463,
                                                     0.647021,1.346249,0.699745,1.412654,2.932174,1.520512,1.494153,3.098146,1.604985,0.809791,1.675544,0.866229,
                                                     0.810192,1.673009,0.863237,1.739711,3.58665 ,1.847753,1.82126 ,3.752188,1.931741,0.973081,2.001861,1.029173});
    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;
}
#endif //LIBND4J_CONVOLUTIONTESTS2_H