Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
contraction_scale_xdl_fp32.cpp 20 KB
Newer Older
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
1
2
3
4
5
6
7
8
9
10 
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.

#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>

#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
Adam Osewski's avatar
Refactor device op implementations into `impl` subdirectory. (#420)  
Adam Osewski committed
11 
#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
12
13
14 
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"

#include "ck/library/utility/check_err.hpp"
Chao Liu's avatar
Clean up conv example, Instances, profiler and test (#324)  
Chao Liu committed
15
16
17 
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp"
Po Yen Chen's avatar
Introduce ck::accumulate_n() (#439)  
Po Yen Chen committed
18 
#include "ck/library/utility/numeric.hpp"
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
19
Bartlomiej Kocot's avatar
Add contraction profiler and tests  
Bartlomiej Kocot committed
20
21 
#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46 
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;

using F32 = float;

using PassThrough = ck::tensor_operation::element_wise::PassThrough;

using ADataType        = F32;
using BDataType        = F32;
using AccDataType      = F32;
using CShuffleDataType = F32;
using DsDataType       = ck::Tuple<>;
using EDataType        = F32;

static constexpr ck::index_t NumDimM = 2;
static constexpr ck::index_t NumDimN = 2;
static constexpr ck::index_t NumDimK = 2;

using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
using CDEElementOp = ck::tensor_operation::element_wise::Scale;

static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;

// clang-format off
Po Yen Chen's avatar
GEMM pipeline v2 (#317)  
Po Yen Chen committed
47 
using DeviceOpInstanceKKN = ck::tensor_operation::device::
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
48
49
50
51
52
53 
        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
Po Yen Chen's avatar
GEMM pipeline v2 (#317)  
Po Yen Chen committed
54 
using DeviceOpInstanceKNN = ck::tensor_operation::device::
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
55
56
57
58
59
60 
        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
Po Yen Chen's avatar
GEMM pipeline v2 (#317)  
Po Yen Chen committed
61 
using DeviceOpInstanceMKN = ck::tensor_operation::device::
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
62
63
64
65
66
67 
        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
Po Yen Chen's avatar
GEMM pipeline v2 (#317)  
Po Yen Chen committed
68 
using DeviceOpInstanceMNN = ck::tensor_operation::device::
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
69
70
71
72
73
74
75 
        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
// clang-format on
Po Yen Chen's avatar
GEMM pipeline v2 (#317)  
Po Yen Chen committed
76 
using DeviceOpInstance = DeviceOpInstanceKKN;
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130 

int main(int argc, char* argv[])
{
    bool do_verification = true;
    int init_method      = 1;
    bool time_kernel     = false;

    // A[M0, M1, K0, K1]
    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
    // B[N0, N1, K0, K1]
    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
    // E[M0, M1, N0, N1]
    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};

    float scale = 1.f;

    if(argc == 1)
    {
        // use default case
    }
    else if(argc == 4)
    {
        do_verification = std::stoi(argv[1]);
        init_method     = std::stoi(argv[2]);
        time_kernel     = std::stoi(argv[3]);
    }
    else if(argc == 23)
    {
        do_verification = std::stoi(argv[1]);
        init_method     = std::stoi(argv[2]);
        time_kernel     = std::stoi(argv[3]);

        const ck::index_t M0 = std::stoi(argv[4]);
        const ck::index_t M1 = std::stoi(argv[5]);

        const ck::index_t N0 = std::stoi(argv[6]);
        const ck::index_t N1 = std::stoi(argv[7]);

        const ck::index_t K0 = std::stoi(argv[8]);
        const ck::index_t K1 = std::stoi(argv[9]);

        a_ms_ks_lengths = {M0, M1, K0, K1};
        a_ms_ks_strides = {
            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};

        b_ns_ks_lengths = {N0, N1, K0, K1};
        b_ns_ks_strides = {
            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};

        e_ms_ns_lengths = {M0, M1, N0, N1};
        e_ms_ns_strides = {
Chao Liu's avatar
Clean up conv example, Instances, profiler and test (#324)  
Chao Liu committed
131 
            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
132
Chao Liu's avatar
Clean up conv example, Instances, profiler and test (#324)  
Chao Liu committed
133 
        scale = std::stof(argv[22]);
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
134
135
136
137
138
139 
    }
    else
    {
        printf("arg1: verification (0=no, 1=yes)\n");
        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
        printf("arg3: time kernel (0=no, 1=yes)\n");
Chao Liu's avatar
Clean up conv example, Instances, profiler and test (#324)  
Chao Liu committed
140 
        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
141
142
143
144
145
146
147 
        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
        printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
        printf("arg22: scale\n");
        exit(0);
    }
Po Yen Chen's avatar
Rangify constructor of HostTensorDescriptor & Tensor<> (#445)  
Po Yen Chen committed
148
149
150
151 
    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169 

    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;

    switch(init_method)
    {
    case 0: break;
    case 1:
        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
        break;
    default:
        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
        break;
    }
Chao Liu's avatar
Clean up conv example, Instances, profiler and test (#324)  
Chao Liu committed
170
171
172 
    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211 

    a_device_buf.ToDevice(a_ms_ks.mData.data());
    b_device_buf.ToDevice(b_ns_ks.mData.data());

    // set zero
    e_device_buf.SetZero();

    auto a_element_op   = AElementOp{};
    auto b_element_op   = BElementOp{};
    auto cde_element_op = CDEElementOp{scale};

    // device operation
    auto op       = DeviceOpInstance{};
    auto invoker  = op.MakeInvoker();
    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
                                    b_device_buf.GetDeviceBuffer(),
                                    std::array<const void*, 0>{},
                                    e_device_buf.GetDeviceBuffer(),
                                    a_ms_ks_lengths,
                                    a_ms_ks_strides,
                                    b_ns_ks_lengths,
                                    b_ns_ks_strides,
                                    std::array<std::vector<ck::index_t>, 0>{},
                                    std::array<std::vector<ck::index_t>, 0>{},
                                    e_ms_ns_lengths,
                                    e_ms_ns_strides,
                                    a_element_op,
                                    b_element_op,
                                    cde_element_op);

    if(!op.IsSupportedArgument(argument))
    {
        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;

        return 0;
    }

    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
Po Yen Chen's avatar
Introduce ck::accumulate_n() (#439)  
Po Yen Chen committed
212
213 
    ck::index_t M =
        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
214
Po Yen Chen's avatar
Introduce ck::accumulate_n() (#439)  
Po Yen Chen committed
215
216 
    ck::index_t N = ck::accumulate_n<ck::index_t>(
        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
217
Po Yen Chen's avatar
Introduce ck::accumulate_n() (#439)  
Po Yen Chen committed
218
219 
    ck::index_t K = ck::accumulate_n<ck::index_t>(
        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235 

    std::size_t flop = std::size_t(2) * M * N * K;
    std::size_t num_btype =
        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;

    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;

    float gb_per_sec = num_btype / 1.E6 / ave_time;

    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
              << op.GetTypeString() << std::endl;

    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());

    if(do_verification)
    {
Bartlomiej Kocot's avatar
Add contraction profiler and tests  
Bartlomiej Kocot committed
236
237
238
239
240
241
242
243
244
245
246 
        using ReferenceOpInstance =
            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
                                                                      NumDimN,
                                                                      NumDimK,
                                                                      ADataType,
                                                                      BDataType,
                                                                      CShuffleDataType,
                                                                      AccDataType,
                                                                      AElementOp,
                                                                      BElementOp,
                                                                      CDEElementOp>;
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
247
248
249
250 

        auto ref_gemm    = ReferenceOpInstance{};
        auto ref_invoker = ref_gemm.MakeInvoker();
Bartlomiej Kocot's avatar
Add contraction profiler and tests  
Bartlomiej Kocot committed
251
252
253
254
255
256
257
258 
        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
        auto ref_argument = ref_gemm.MakeArgument(a_ms_ks,
                                                  b_ns_ks,
                                                  empty_tensor,
                                                  e_ms_ns_host_result,
                                                  a_element_op,
                                                  b_element_op,
                                                  cde_element_op);
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
259
260
261 

        ref_invoker.Run(ref_argument);
Po Yen Chen's avatar
Rangify constructor of HostTensorDescriptor & Tensor<> (#445)  
Po Yen Chen committed
262 
        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
Chao Liu's avatar
N-D Tensor Contraction example, instance, and client example (#270)  
Chao Liu committed
263
264
265
266 
    }

    return 0;
}