merge from develop and revisison for pr#881

example/01_gemm/common.hpp

@@ -33,6 +33,19 @@ struct ProblemSize final
     ck::index_t StrideC = 4096;
 };
 
+struct ProblemSizeStreamK final
+{
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideC = 4096;
+
+    ck::index_t NumSKBlocks = -1;
+};
+
 struct ExecutionConfig final
 {
     bool do_verification = true;
@@ -48,8 +61,17 @@ using Col = ck::tensor_layout::gemm::ColumnMajor;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-inline bool
-parse_cmd_args(int argc, char* argv[], ProblemSize& problem_size, ExecutionConfig& config)
+template <typename ProblemType>
+bool parse_cmd_args(int, char*[], ProblemType&, ExecutionConfig&)
+{
+    return false;
+}
+
+template <>
+bool parse_cmd_args<ProblemSize>(int argc,
+                                 char* argv[],
+                                 ProblemSize& problem_size,
+                                 ExecutionConfig& config)
 {
     if(argc == 1)
     {
@@ -87,3 +109,52 @@ parse_cmd_args(int argc, char* argv[], ProblemSize& problem_size, ExecutionConfi
 
     return true;
 }
+
+template <>
+bool parse_cmd_args<ProblemSizeStreamK>(int argc,
+                                        char* argv[],
+                                        ProblemSizeStreamK& problem_size,
+                                        ExecutionConfig& config)
+{
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc >= 10)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+
+        problem_size.M = std::stoi(argv[4]);
+        problem_size.N = std::stoi(argv[5]);
+        problem_size.K = std::stoi(argv[6]);
+
+        problem_size.StrideA = std::stoi(argv[7]);
+        problem_size.StrideB = std::stoi(argv[8]);
+        problem_size.StrideC = std::stoi(argv[9]);
+
+        if(argc >= 11)
+        {
+            problem_size.NumSKBlocks = std::stoi(argv[10]);
+        }
+    }
+    else
+    {
+        std::cerr << "arg1: verification (0=no, 1=yes)" << std::endl
+                  << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)"
+                  << std::endl
+                  << "arg3: time kernel (0=no, 1=yes)" << std::endl
+                  << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC" << std::endl
+                  << "arg10: NumSKBlocks(optional)" << std::endl;
+        return false;
+    }
+
+    return true;
+}

example/01_gemm/gemm_dpp_fp16.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+
+using ADataType   = ck::half_t;
+using BDataType   = ck::half_t;
+using AccDataType = float;
+using CDataType   = ck::half_t;
+
+using F16 = ck::half_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDpp
+// ######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+// ######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise|Spacialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+// ######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+// ######|          |          |          |            |        |        |        |            |            |            |              |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+         < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,   GemmDefault,   128,    64,    64,    64,   8,   2,   32,    8,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               5,               1>;
+// // clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_xdl_bf16_rtn.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/utility/type_convert.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+
+using ADataType        = ck::bhalf_t;
+using BDataType        = ck::bhalf_t;
+using CDataType        = ck::bhalf_t;
+using AccDataType      = float;
+using CShuffleDataType = float;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = ck::tensor_operation::element_wise::ConvertBF16RTN;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|           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|        Type|         DataType| 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|
+// ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    32,   8,   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<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_xdl_f8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+
+using ADataType        = ck::f8_t;
+using BDataType        = ck::f8_t;
+using CDataType        = ck::f8_t;
+using AccDataType      = float;
+using CShuffleDataType = ck::f8_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|           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|        Type|         DataType| 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|
+// ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 64, 1, 4>,              16>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_xdl_fp16_f8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+
+using ADataType        = ck::f8_t;
+using BDataType        = ck::half_t;
+using CDataType        = ck::half_t;
+using AccDataType      = float;
+using CShuffleDataType = ck::half_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto LoopSched   = ck::make_default_loop_scheduler();
+static constexpr auto PipelineVer = ck::PipelineVersion::v1;
+using ComputeType                 = ck::half_t;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|           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|      Loop|    Pipeline|  ComputeType|
+// ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| 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| Scheduler|     Version|             |
+// ######|        |        |        |          |          |          |            |                 |   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|          |            |             |
+// ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |          |            |             |
+        < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    32,   8,   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<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,  LoopSched, PipelineVer, ComputeType>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }

example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp

@@ -204,9 +204,9 @@ int main(int argc, char* argv[])
 
     if(!gemm.IsSupportedArgument(argument))
     {
-        throw std::runtime_error(
-            "wrong! device_gemm with the specified compilation parameters does "
-            "not support this GEMM problem");
+        std::cerr << gemm.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
     }
 
     float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});

example/01_gemm/gemm_xdl_streamk.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#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 BLayout = Col;
+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|
+// ######|          |          |          |            |        |        |        |            |            |            |       |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+    < 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>;
+
+
+
+// // clang-format on
+// clang-format on
+
+using DeviceGemmInstance = DeviceGemmStreamK;
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_streamk_example(argc, argv); }

example/01_gemm/gemm_xdl_wavelet_fp16.cpp

@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp"
 
 using ADataType        = ck::half_t;
 using BDataType        = ck::half_t;

example/01_gemm/run_gemm_example.inc

@@ -3,7 +3,10 @@
 
 #pragma once
 
-bool run_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
+#include "ck/tensor_operation/gpu/device/device_gemm_streamk.hpp"
+
+template <typename ProblemType>
+bool run_gemm(const ProblemType& problem_size, const ExecutionConfig& config)
 {
 #if defined(BUILD_INT4_EXAMPLE) && defined(CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4)
     static_assert(sizeof(ck::int4_t) == sizeof(int8_t));
@@ -11,7 +14,12 @@ bool run_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
 
     using namespace ck::literals;
 
-    auto& [M, N, K, StrideA, StrideB, StrideC] = problem_size;
+    auto M       = problem_size.M;
+    auto N       = problem_size.N;
+    auto K       = problem_size.K;
+    auto StrideA = problem_size.StrideA;
+    auto StrideB = problem_size.StrideB;
+    auto StrideC = problem_size.StrideC;
 
     auto f_host_tensor_descriptor =
         [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
@@ -25,12 +33,37 @@ bool run_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
             }
         };
 
+    auto f_get_default_stride =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            if(stride == 0)
+            {
+                // give a chance if stride is zero, return a default packed stride
+                if constexpr(std::is_same_v<decltype(layout), ck::tensor_layout::gemm::RowMajor>)
+                {
+                    return col;
+                }
+                else
+                {
+                    return row;
+                }
+            }
+            else
+                return stride;
+        };
+
+    StrideA = f_get_default_stride(M, K, StrideA, ALayout{});
+    StrideB = f_get_default_stride(K, N, StrideB, BLayout{});
+    StrideC = f_get_default_stride(M, N, StrideC, CLayout{});
+
     Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
     Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
 
     switch(config.init_method)
     {
-    case 0: break;
+    case 0:
+        ck::utils::FillConstant<ADataType>{static_cast<ADataType>(1.f)}(a_m_k);
+        ck::utils::FillConstant<BDataType>{static_cast<BDataType>(1.f)}(b_k_n);
+        break;
     case 1:
         ck::utils::FillUniformDistributionIntegerValue<ADataType>{-5.f, 5.f}(a_m_k);
         ck::utils::FillUniformDistributionIntegerValue<BDataType>{-5.f, 5.f}(b_k_n);
@@ -66,43 +99,115 @@ bool run_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
     a_m_k_device_buf.ToDevice(a_m_k.mData.data());
     b_k_n_device_buf.ToDevice(b_k_n.mData.data());
 #endif
+    DeviceMem workspace;
 
     auto a_element_op = AElementOp{};
     auto b_element_op = BElementOp{};
     auto c_element_op = CElementOp{};
 
+    using BaseStreamK = ck::tensor_operation::device::DeviceGemmStreamK<ALayout,
+                                                                        BLayout,
+                                                                        CLayout,
+                                                                        ADataType,
+                                                                        BDataType,
+                                                                        CDataType,
+                                                                        AElementOp,
+                                                                        BElementOp,
+                                                                        CElementOp>;
+
     // do GEMM
-    auto gemm     = DeviceGemmInstance{};
-    auto invoker  = gemm.MakeInvoker();
-    auto argument = gemm.MakeArgument(
+    auto gemm      = DeviceGemmInstance{};
+    auto invoker   = gemm.MakeInvoker();
+    float ave_time = 0;
+
+    if constexpr(std::is_same<ProblemType, ProblemSize>::value &&
+                 !std::is_base_of<BaseStreamK, DeviceGemmInstance>::value)
+    {
+        auto argument = gemm.MakeArgument(
 #ifdef BUILD_INT4_EXAMPLE
-        static_cast<KernelADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-        static_cast<KernelBDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-        static_cast<KernelCDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+            static_cast<KernelADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+            static_cast<KernelBDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+            static_cast<KernelCDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
 #else
-        static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-        static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-        static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+            static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+            static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+            static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
 #endif
-        M,
-        N,
-        K,
-        StrideA,
-        StrideB,
-        StrideC,
-        a_element_op,
-        b_element_op,
-        c_element_op);
-
-    if(!gemm.IsSupportedArgument(argument))
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            StrideC,
+            a_element_op,
+            b_element_op,
+            c_element_op);
+
+        if(!gemm.IsSupportedArgument(argument))
+        {
+            std::cerr << gemm.GetTypeString() << " does not support this problem" << std::endl;
+
+            return true;
+        }
+
+        ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+    }
+    else if constexpr(std::is_same<ProblemType, ProblemSizeStreamK>::value &&
+                      std::is_base_of<BaseStreamK, DeviceGemmInstance>::value)
     {
-        std::cerr << gemm.GetTypeString() << " does not support this problem" << std::endl;
-
-        return true;
+        auto argument = gemm.MakeArgument(
+#ifdef BUILD_INT4_EXAMPLE
+            static_cast<KernelADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+            static_cast<KernelBDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+            static_cast<KernelCDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+#else
+            static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+            static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+            static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+#endif
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            StrideC,
+            a_element_op,
+            b_element_op,
+            c_element_op,
+            problem_size.NumSKBlocks);
+
+        if(!gemm.IsSupportedArgument(argument))
+        {
+            std::cerr << gemm.GetTypeString() << " does not support this problem" << std::endl;
+
+            return true;
+        }
+
+        std::size_t workspace_size = gemm.GetWorkSpaceSize(&argument);
+        if(workspace_size != 0)
+        {
+            workspace.Realloc(workspace_size);
+            gemm.SetWorkSpacePointer(&argument, workspace.GetDeviceBuffer());
+        }
+
+        ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+
+#if 0
+        // TODO!!!!!
+        if(workspace_size != 0){
+            float * ws_ptr = reinterpret_cast<float*>(malloc(workspace_size));
+            size_t ws_dwords = workspace_size / sizeof(float);
+            workspace.FromDevice(ws_ptr);
+
+            for(size_t i = 0; i < ws_dwords; i++) {
+                uint32_t rere = reinterpret_cast<uint32_t*>(ws_ptr)[i];
+                printf("%4lu : %f(0x%08x)\n", i, ws_ptr[i], rere);
+            }
+            free(ws_ptr);
+        }
+#endif
     }
 
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
-
     std::size_t flop = 2_uz * M * N * K;
     std::size_t num_btype =
         sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(CDataType) * M * N;
@@ -149,3 +254,11 @@ bool run_gemm_example(int argc, char* argv[])
 
     return !parse_cmd_args(argc, argv, problem_size, config) || run_gemm(problem_size, config);
 }
+
+bool run_gemm_streamk_example(int argc, char* argv[])
+{
+    ProblemSizeStreamK problem_size;
+    ExecutionConfig config;
+
+    return !parse_cmd_args(argc, argv, problem_size, config) || run_gemm(problem_size, config);
+}