Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
gemm_xdl_streamk.cpp 6.1 KB
Newer Older
carlushuang's avatar
initial stream-k implementation with example (#699)  
carlushuang committed
1 
// SPDX-License-Identifier: MIT
Rostyslav Geyyer's avatar
Add a gpu gemm reference kernel (#1528)  
Rostyslav Geyyer committed
2 
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
carlushuang's avatar
initial stream-k implementation with example (#699)  
carlushuang committed
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29 

#include "common.hpp"

#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp"

using ADataType        = ck::half_t;
using BDataType        = ck::half_t;
using AccDataType      = float;
using CShuffleDataType = float;
using CDataType        = ck::half_t;

using F16 = ck::half_t;

using ALayout = Row;
using BLayout = Row;
using CLayout = Row;

using AElementOp = PassThrough;
using BElementOp = PassThrough;
using CElementOp = PassThrough;

// clang-format off
using DeviceGemmStreamK = ck::tensor_operation::device::DeviceGemmXdlStreamK
// ######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|  Block|  MPer|  NPer| K0Per| K1| 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|        Type|        |        |        | Elementwise| Elementwise| Elementwise|   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|       |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
// ######|          |          |          |            |        |        |        |            |            |            |       |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
jefyang1's avatar
Fix test_gemm_universal test_grouped_convnd_fwd and example_gemm_xdl_streamk on gfx950 (#1891)  
jefyang1 committed
30
31
32 
#if defined(CK_USE_AMD_MFMA_GFX950)
    < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    256,   256,   128,     4, 16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,      S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>;
#else // defined(CK_USE_AMD_MFMA_GFX950)
carlushuang's avatar
initial stream-k implementation with example (#699)  
carlushuang committed
33
34
35
36
37 
    < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,      S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>;

          // < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    256,   256,   128,     4,   8,  32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              2,         1,           1,           1,               S<1, 32, 1, 8>,              8>;
      //  < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    128,   32,   64,     4,   8,  32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>;
       // < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    128,   32,   128,     4,   8,  32,   32,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              2,         1,           1,           1,               S<1, 32, 1, 4>,              8>;
jefyang1's avatar
Fix test_gemm_universal test_grouped_convnd_fwd and example_gemm_xdl_streamk on gfx950 (#1891)  
jefyang1 committed
38 
#endif // defined(CK_USE_AMD_MFMA_GFX950)
carlushuang's avatar
initial stream-k implementation with example (#699)  
carlushuang committed
39
40
41
42
43
44
45
46
47
48
49 



// // clang-format on
// clang-format on

using DeviceGemmInstance = DeviceGemmStreamK;

using ReferenceGemmInstance = ck::tensor_operation::host::
    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
Rostyslav Geyyer's avatar
Add a gpu gemm reference kernel (#1528)  
Rostyslav Geyyer committed
50
51
52
53
54
55
56
57
58
59
60 
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
                                                                             BLayout,
                                                                             CLayout,
                                                                             ADataType,
                                                                             BDataType,
                                                                             CDataType,
                                                                             AccDataType,
                                                                             AElementOp,
                                                                             BElementOp,
                                                                             CElementOp>;
carlushuang's avatar
initial stream-k implementation with example (#699)  
carlushuang committed
61
62
63 
#include "run_gemm_example.inc"

int main(int argc, char* argv[]) { return !run_gemm_streamk_example(argc, argv); }