[GEMM] Gemm universal device operation (#1154)

* Optimize GEMM on MI200/300:
1. Add new blockwise gemm pipeline
2. Add irregular splitk intances

* clang format + typo fix

* Fix a bug

* initial commit

* Add more instances to irregular splitk

* blkgemm pipeline v1~4 prototype

* Sanity Checked. Known issue:
1. Poor performance of splitk
2. Register spill on blkgemmpipeline v3

* Sanity and Performance fix:
1. fix a bug related to sanity in grouped b2c mapping
2. fix a bug related to sanity and performance in splitk offset

* Sanity and API update:
1. Remove prefetch stage
2. Fix valid check bug
3, Add first gemm_universal instance into ckProfiler

* Add NN instances for gemm universal

* 1. Add NT instances for gemm_universal
2. Fix a bug about Kpadding in gemm_universal

* Fix a bug regarding padding Odd K number

* remove kernel print

* Fix KPadding bug...

* Update safety check

* another try to fix kpadding..

* Sanity checked

* new instances..

* clang format+typo fix

* remove clang format script's change

* Add non-hotloop compile option

* 1. Add fp16xfp8 example
2. pull packed convert f8 from pr1150

* Some miscs.. opt and fix

* Add pipeline description docs

* Split universal gemm instance library to cut profiler compiling time

* uncomment cmakefile

* Fix a bug caused by blockwise_gemm_pipe_v2

* reduce default splitk to 1

* Add 224x256x64 tile size

* update, including:
1. Experiment pipeline 5~7
2. Optimization for pipeline 4
3. Organized instance library

* temp save

* temp save

* Permuted lds layout, sanity and function checked

* clang format

* Move OOB check from RunRead to RunWrite, for better software pipeline.
TODO: agpr spill when NN layout

* clangformat

* A/B splitpipe scheduler for v3

* Fix two bugs

* bug fix

* fix a bug in oob check

* Example for mixed fp16_fp8 gemm

* Clean experimental code blocks

* Add mixed precision gemm into profiler

* tempsave

* optimize m/n major lds layout

* Add RRR GEMM  mixed precision instances

* Optimize f8 matrix transpose

* Add test_gemm_universal

* A/B spilt schedule for blkpip v5

* Take ds_read2 into iglp scheduling scheme

* format

* fixed cmake

* Add llvm-option into CI cmake flag

---------
Co-authored-by: Jing Zhang <jizhan@amd.com>

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_comp_mnpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_comp_mnpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_comp_instances<GemmMNPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v1_default_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v1_default_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_instances<Intrawave, GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v1_kpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_instances<Intrawave, GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v1_mnkpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v1_mnkpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_instances<Intrawave, GemmMNKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v2_default_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v2_default_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_instances<Interwave, GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v2_kpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v2_kpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_instances<Interwave, GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v2_mnkpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_v2_mnkpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_kn_mn_mem_instances<Interwave, GemmMNKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_v3.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = f8_t;
+using F16 = half_t;
+using F32 = float;
+
+using Row = tensor_layout::gemm::RowMajor;
+using Col = tensor_layout::gemm::ColumnMajor;
+
+template <index_t... Is>
+using S = Sequence<Is...>;
+
+using PassThrough = element_wise::PassThrough;
+
+static constexpr auto GemmDefault    = GemmSpecialization::Default;
+static constexpr auto GemmKPadding   = GemmSpecialization::KPadding;
+static constexpr auto GemmMNPadding  = GemmSpecialization::MNPadding;
+static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
+
+static constexpr auto Intrawave = BlockGemmPipelineScheduler::Intrawave;
+static constexpr auto Interwave = BlockGemmPipelineScheduler::Interwave;
+
+template <GemmSpecialization GemmSpec>
+using device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_instances = std::tuple<
+    // clang-format off
+        //#########################| ALayout| BLayout| CLayout|AData| BData| CData| AccData| Cshuffle|           A|           B|           C|          GEMM| 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|                         Block-wiseGemm|               Block-wiseGemm|
+        //#########################|        |        |        | Type|  Type|  Type|    Type|     Type| Elementwise| Elementwise| Elementwise|Specialization|  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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|                               Pipeline|                     Pipeline|
+        //#########################|        |        |        |     |      |      |        |         |   Operation|   Operation|   Operation|              |      |      |      |      |    |    |    |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|                              Scheduler|                     Verision|
+        //#########################|        |        |        |     |      |      |        |         |            |            |            |              |      |      |      |      |    |    |    |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |                                       |                             |
+        
+        // Compute friendly
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,  16,   8,  32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   224,    64,  16,   8,  16,   16,    8,    7,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          2,           1,                   S<1, 64, 1, 4>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,  16,   8,  32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v3>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,  16,   8,  32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v5>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   128,   128,    64,  16,   8,  32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               8,  BlockGemmPipelineScheduler::Interwave, BlockGemmPipelineVersion::v1>
+    // clang-format on
+    >;
+
+template <BlockGemmPipelineScheduler BlkGemmPipeSched, GemmSpecialization GemmSpec>
+using device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_instances = std::tuple<
+    // clang-format off
+        //#########################| ALayout| BLayout| CLayout|AData| BData| CData| AccData| Cshuffle|           A|           B|           C|          GEMM| 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|    Block-wiseGemm|               Block-wiseGemm|
+        //#########################|        |        |        | Type|  Type|  Type|    Type|     Type| Elementwise| Elementwise| Elementwise|Specialization|  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_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|          Pipeline|                     Pipeline|
+        //#########################|        |        |        |     |      |      |        |         |   Operation|   Operation|   Operation|              |      |      |      |      |    |    |    |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|         Scheduler|                     Verision|
+        //#########################|        |        |        |     |      |      |        |         |            |            |            |              |      |      |      |      |    |    |    |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |                  |                             |
+
+        // Latency friendly 
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    32,   16,   128,   16,   8,  16,   16,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               2,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,    64,    16,   16,   128,   16,   8,  16,   16,    1,    1,     S<8, 8, 1>,      S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 4, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 4>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,    64,    16,   16,   128,   16,   8,  16,   16,    1,    1,     S<8, 8, 1>,      S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 4, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 4>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   32,   128,   16,   8,  16,   16,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v1>,
+        // Memory friendly
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   32,   128,   16,   8,  32,   32,    2,    1,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 16, 1>,    S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   256,   256,   16,   128,   16,   8,  16,   16,    4,    1,     S<8, 32, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 32, 1, 8>,               2,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,   128,   32,   128,   16,   8,  32,   32,    2,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,   128,   16,   128,   16,   8,  16,   16,    4,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               2,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    64,   32,   128,   16,   8,  32,   32,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    64,   16,   128,   16,   8,  16,   16,    2,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               2,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    32,   16,   128,   16,   8,  16,   16,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               2,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,    64,    16,   16,   128,   16,   8,  16,   16,    1,    1,     S<8, 8, 1>,      S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 4, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 4>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,    64,    16,   16,   128,   16,   8,  16,   16,    1,    1,     S<8, 8, 1>,      S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 4, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 4>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   32,   128,   16,   8,  16,   16,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,   64,   128,   16,   8,  16,   16,    1,    2,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    32,   64,   128,   16,   8,  32,   32,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               8,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    16,  128,   128,   16,   8,  16,   16,    1,    4,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               4,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffleV3<  Row,     Col,     Row,      F8,   F16,  F16,   F32,     F16,      PassThrough, PassThrough, PassThrough,       GemmSpec,   128,    32,  128,   128,   16,   8,  32,   32,    1,    2,     S<8, 16, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 8, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,              8,              8,          0,          1,           1,                   S<1, 16, 1, 8>,               8,  BlkGemmPipeSched, BlockGemmPipelineVersion::v2>
+    // clang-format on
+    >;
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_default_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_default_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_instances<GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_kpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_kpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_instances<GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_mnkpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_mnkpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_instances<GemmMNKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_mnpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_mnpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_comp_instances<GemmMNPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v1_default_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v1_default_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_instances<Intrawave, GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v1_kpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v1_kpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_instances<Intrawave, GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v1_mnkpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v1_mnkpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_instances<Intrawave, GemmMNKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v2_default_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v2_default_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_instances<Interwave, GemmDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v2_kpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v2_kpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_instances<Interwave, GemmKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/gemm_universal/device_gemm_xdl_universal_f8_f16_f16/device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v2_mnkpadding_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_v2_mnkpadding_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmV2<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_xdl_universal_f8_f16_f16_mk_nk_mn_mem_instances<Interwave, GemmMNKPadding>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

profiler/include/profiler/profile_gemm_universal_impl.hpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+#include <iostream>
+#include <typeinfo>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_v3.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_universal.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CDataType,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout>
+bool profile_gemm_universal_impl(int do_verification,
+                                 int init_method,
+                                 bool do_log,
+                                 bool time_kernel,
+                                 int M,
+                                 int N,
+                                 int K,
+                                 int StrideA,
+                                 int StrideB,
+                                 int StrideC,
+                                 int KBatch,
+                                 int n_warmup,
+                                 int n_iter)
+{
+    bool pass = true;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(is_same<decltype(layout), tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    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{}));
+    Tensor<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
+    Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "c_m_n: " << c_m_n_device_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-1, 2});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-1, 2});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+    }
+
+    using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using CElementOp = ck::tensor_operation::element_wise::PassThrough;
+
+    const auto a_element_op = AElementOp{};
+    const auto b_element_op = BElementOp{};
+    const auto c_element_op = CElementOp{};
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem c_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmV2<ALayout,
+                                                                BLayout,
+                                                                CLayout,
+                                                                ADataType,
+                                                                BDataType,
+                                                                CDataType,
+                                                                AElementOp,
+                                                                BElementOp,
+                                                                CElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    // Run reference GEMM
+    if(do_verification)
+    {
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                CDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                BElementOp,
+                                                                                CElementOp>;
+
+        auto ref_gemm    = ReferenceGemmInstance{};
+        auto ref_invoker = ref_gemm.MakeInvoker();
+
+        auto ref_argument = ref_gemm.MakeArgument(
+            a_m_k, b_k_n, c_m_n_host_result, a_element_op, b_element_op, c_element_op);
+
+        ref_invoker.Run(ref_argument);
+    }
+
+    std::string best_op_name;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+    float best_kbatch     = 0;
+
+    // profile device GEMM instances
+    for(auto& op_ptr : op_ptrs)
+    {
+        std::vector<int> kbatch_list = {1, 2, 4, 8, 12, 16, 19, 20, 32, 38};
+
+        if(KBatch > 0)
+        {
+            kbatch_list = {KBatch};
+        }
+
+        for(std::size_t i = 0; i < kbatch_list.size(); i++)
+        {
+            auto kbatch_curr = kbatch_list[i];
+
+            auto argument_ptr =
+                op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+                                            static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
+                                            static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+                                            M,
+                                            N,
+                                            K,
+                                            StrideA,
+                                            StrideB,
+                                            StrideC,
+                                            kbatch_curr,
+                                            a_element_op,
+                                            b_element_op,
+                                            c_element_op);
+
+            auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+            if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+            {
+
+                // re-init C to zero before profiling next kernel
+                c_device_buf.SetZero();
+
+                invoker_ptr->Run(argument_ptr.get(),
+                                 StreamConfig{nullptr, false, 0, n_warmup, n_iter});
+
+                if(do_verification)
+                {
+                    c_device_buf.FromDevice(c_m_n_device_result.mData.data());
+
+                    pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
+
+                    if(do_log)
+                    {
+                        LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
+                        LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
+                        LogRangeAsType<float>(
+                            std::cout << "c_host  : ", c_m_n_host_result.mData, ",")
+                            << std::endl;
+                        LogRangeAsType<float>(
+                            std::cout << "c_device: ", c_m_n_device_result.mData, ",")
+                            << std::endl;
+                    }
+                }
+
+                std::string op_name = op_ptr->GetTypeString();
+
+                float ave_time = invoker_ptr->Run(
+                    argument_ptr.get(), StreamConfig{nullptr, time_kernel, 0, n_warmup, n_iter});
+
+                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(CDataType) * 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: " << std::setw(10) << ave_time << " ms, " << tflops
+                          << " TFlops, " << gb_per_sec << " GB/s, " << op_name << ", KBatch "
+                          << kbatch_curr << std::endl;
+
+#if defined CK_ENABLE_FP8
+                // set softer tolerances for fp8
+                if constexpr(is_same_v<ADataType, f8_t> || is_same_v<BDataType, f8_t> ||
+                             is_same_v<CDataType, f8_t>)
+                {
+                    std::string msg = "Error: Incorrect results!";
+                    double rtol     = 1e-1;
+                    double atol     = 1e-1;
+                    pass            = pass & ck::utils::check_err(
+                                      c_m_n_device_result, c_m_n_host_result, msg, rtol, atol);
+                }
+                else
+                {
+#endif
+                    pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
+#if defined CK_ENABLE_FP8
+                }
+#endif
+
+                if(tflops > best_tflops)
+                {
+                    best_op_name    = op_name;
+                    best_tflops     = tflops;
+                    best_ave_time   = ave_time;
+                    best_gb_per_sec = gb_per_sec;
+                    best_kbatch     = kbatch_curr;
+                }
+            }
+            else
+            {
+                std::cout << op_ptr->GetTypeString() << " does not support this problem"
+                          << std::endl;
+            }
+        }
+    }
+
+    if constexpr(is_same<CDataType, float>::value)
+    {
+        std::cout << "Best Perf for datatype = f32";
+    }
+    else if constexpr(is_same<CDataType, half_t>::value)
+    {
+        std::cout << "Best Perf for datatype = f16";
+    }
+    else if constexpr(is_same<CDataType, bhalf_t>::value)
+    {
+        std::cout << "Best Perf for datatype = bf16";
+    }
+    else if constexpr(is_same<CDataType, int8_t>::value)
+    {
+        std::cout << "Best Perf for datatype = int8";
+    }
+
+    if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " ALayout =  RowMajor";
+    }
+    else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " ALayout =  ColumnMajor";
+    }
+
+    if constexpr(is_same<BLayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " BLayout =  RowMajor";
+    }
+    else if constexpr(is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " BLayout =  ColumnMajor";
+    }
+
+    std::cout << " M = " << M << " N = " << N << " K = " << K << " StrideA = " << StrideA
+              << " StrideB = " << StrideB << " StrideC = " << StrideC << " KBatch = " << best_kbatch
+              << " : " << best_ave_time << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec
+              << " GB/s, " << best_op_name << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck

profiler/src/CMakeLists.txt

@@ -49,6 +49,7 @@ if(GPU_TARGETS MATCHES "gfx9")
   list(APPEND PROFILER_SOURCES profile_gemm_add_multiply.cpp)
   list(APPEND PROFILER_SOURCES profile_gemm_bias_add_reduce.cpp)
   list(APPEND PROFILER_SOURCES profile_gemm_splitk.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_universal.cpp)
   list(APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu.cpp)
   list(APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu_add.cpp)
   list(APPEND PROFILER_SOURCES profile_conv_bwd_data.cpp)
@@ -115,6 +116,7 @@ if(GPU_TARGETS MATCHES "gfx9")
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_reduce_instance)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_multiply_add_instance)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_universal_instance)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance)