Merge branch 'uif2-initial' into uif2-migraphx

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_lds_direct_load_f32_f32_f32_km_nk_mn_instance.cpp

@@ -32,7 +32,8 @@ using device_gemm_xdl_c_shuffle_lds_direct_load_f32_f32_f32_km_nk_mn_instances =
     // ##################################|        |        |        |  Type|  Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster| SrcAccessOrder|   SrcVectorDim|         Scalar| AddExtraM|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|         Scalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
     // ##################################|        |        |        |      |      |      |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|               |               |      PerVector|          | Lengths_K0_N_K1|               |              |      PerVector|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
     // ##################################|        |        |        |      |      |      |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |          |                |               |              |               |          |            |            |                             |                |
-    DeviceGemm_Xdl_CShuffle_LdsDirectLoad<     Col,     Col,     Row,   F32,   F32,   F32,     F32,      F32, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,    64,    32,   8,   8,   32,   32,    1,    1,      S<4, 8, 8>,     S<0, 2, 1>,              1,              1,         1,      S<4, 8, 8>,     S<1, 0, 2>,             2,              1,         1,           1,           1,                S<1, 8, 1, 8>,               4>
+    DeviceGemm_Xdl_CShuffle_LdsDirectLoad<     Col,     Col,     Row,   F32,   F32,   F32,     F32,      F32, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,    64,    32,   8,   8,   32,   32,    1,    1,      S<4, 8, 8>,     S<0, 2, 1>,              1,              1,         1,      S<4, 8, 8>,     S<1, 0, 2>,             2,              1,         1,           1,           1,                S<1, 8, 1, 8>,               4>,
+    DeviceGemm_Xdl_CShuffle_LdsDirectLoad<     Col,     Col,     Row,   F32,   F32,   F32,     F32,      F32, PassThrough, PassThrough, PassThrough,    GemmDefault,        2,   256,    64,    64,    32,   8,   8,   32,   32,    1,    1,      S<4, 8, 8>,     S<0, 2, 1>,              1,              1,         1,      S<4, 8, 8>,     S<1, 0, 2>,             2,              1,         1,           1,           1,                S<1, 8, 1, 8>,               4>
     // clang-format on
     >;
 

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_lds_direct_load_f32_f32_f32_mk_kn_mn_instance.cpp

@@ -31,7 +31,8 @@ using device_gemm_xdl_c_shuffle_lds_direct_load_f32_f32_f32_mk_kn_mn_instances =
     // ##################################|        |        |        |  Type|  Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster| SrcAccessOrder|   SrcVectorDim|         Scalar| AddExtraM|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|         Scalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
     // ##################################|        |        |        |      |      |      |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|               |               |      PerVector|          | Lengths_K0_N_K1|               |              |      PerVector|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
     // ##################################|        |        |        |      |      |      |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |          |                |               |              |               |          |            |            |                             |                |
-    DeviceGemm_Xdl_CShuffle_LdsDirectLoad<     Row,     Row,     Row,   F32,   F32,   F32,     F32,      F32, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,    64,    32,   8,   8,   32,   32,    1,    1,      S<4, 8, 8>,     S<1, 0, 2>,              2,              1,         1,      S<4, 8, 8>,     S<0, 2, 1>,             1,              1,         1,           1,           1,                S<1, 8, 1, 8>,               4>
+    DeviceGemm_Xdl_CShuffle_LdsDirectLoad<     Row,     Row,     Row,   F32,   F32,   F32,     F32,      F32, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,    64,    32,   8,   8,   32,   32,    1,    1,      S<4, 8, 8>,     S<1, 0, 2>,              2,              1,         1,      S<4, 8, 8>,     S<0, 2, 1>,             1,              1,         1,           1,           1,                S<1, 8, 1, 8>,               4>,
+    DeviceGemm_Xdl_CShuffle_LdsDirectLoad<     Row,     Row,     Row,   F32,   F32,   F32,     F32,      F32, PassThrough, PassThrough, PassThrough,    GemmDefault,        2,   256,    64,    64,    32,   8,   8,   32,   32,    1,    1,      S<4, 8, 8>,     S<1, 0, 2>,              2,              1,         1,      S<4, 8, 8>,     S<0, 2, 1>,             1,              1,         1,           1,           1,                S<1, 8, 1, 8>,               4>
     // clang-format on
     >;
 

library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_lds_direct_load_f32_f32_f32_mk_nk_mn_instance.cpp

@@ -25,7 +25,6 @@ using S = ck::Sequence<Is...>;
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
 
 using device_gemm_xdl_c_shuffle_lds_direct_load_f32_f32_f32_mk_nk_mn_instances = std::tuple<
     // clang-format off
@@ -34,7 +33,7 @@ using device_gemm_xdl_c_shuffle_lds_direct_load_f32_f32_f32_mk_nk_mn_instances =
     // ##################################|        |        |        |      |      |      |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|               |               |      PerVector|          | Lengths_K0_N_K1|               |              |      PerVector|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
     // ##################################|        |        |        |      |      |      |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |          |                |               |              |               |          |            |            |                             |                |
     DeviceGemm_Xdl_CShuffle_LdsDirectLoad<     Row,     Col,     Row,   F32,   F32,   F32,     F32,      F32, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,    64,    32,   8,   8,   32,   32,    1,    1,      S<4, 8, 8>,     S<1, 0, 2>,              2,              1,         1,      S<4, 8, 8>,     S<1, 0, 2>,             2,              1,         1,           1,           1,                S<1, 8, 1, 8>,               4>,
-    DeviceGemm_Xdl_CShuffle_LdsDirectLoad<     Row,     Col,     Row,   F32,   F32,   F32,     F32,      F32, PassThrough, PassThrough, PassThrough,  GemmMNPadding,        1,   256,    64,    64,    32,   8,   8,   32,   32,    1,    1,      S<4, 8, 8>,     S<1, 0, 2>,              2,              1,         1,      S<4, 8, 8>,     S<1, 0, 2>,             2,              1,         1,           1,           1,                S<1, 8, 1, 8>,               4>
+    DeviceGemm_Xdl_CShuffle_LdsDirectLoad<     Row,     Col,     Row,   F32,   F32,   F32,     F32,      F32, PassThrough, PassThrough, PassThrough,    GemmDefault,        2,   256,    64,    64,    32,   8,   8,   32,   32,    1,    1,      S<4, 8, 8>,     S<1, 0, 2>,              2,              1,         1,      S<4, 8, 8>,     S<1, 0, 2>,             2,              1,         1,           1,           1,                S<1, 8, 1, 8>,               4>
     // clang-format on
     >;
 

profiler/include/profiler/profile_gemm_impl.hpp

@@ -166,7 +166,7 @@ int profile_gemm_impl(int do_verification,
             std::string op_name = op_ptr->GetTypeString();
 
             float avg_time =
-                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel, 0, 10, 50});
 
             std::size_t flop = std::size_t(2) * M * N * K;
 

profiler/src/profile_transpose.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_transpose_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+enum struct MatrixLayout
+{
+    NCDHW, // 0
+    NCHWD, // 1
+};
+
+enum struct DataType
+{
+    F32_F32_F32_F32_F32, // 0
+    F16_F16_F16_F16_F16, // 1
+};
+
+#define OP_NAME "transpose"
+#define OP_DESC "Transpose"
+
+int profile_transpose(int argc, char* argv[])
+{
+    if(argc != 15)
+    {
+        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
+        printf("arg2: data type (0: fp32; 1: fp16)\n");
+        // printf("arg3: matrix layout (NCDHW -> NDCHW);\n");
+        printf("arg4: verification (0: no; 1: yes)\n");
+        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
+        printf("arg6: print tensor value (0: no; 1: yes)\n");
+        printf("arg7: time kernel (0=no, 1=yes)\n");
+        printf("arg8 to 13: N, C, D, H, W\n");
+        exit(1);
+    }
+
+    const auto data_type = static_cast<DataType>(std::stoi(argv[2]));
+    // const auto layout          = static_cast<MatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification   = std::stoi(argv[3]);
+    const int init_method        = std::stoi(argv[4]);
+    const bool do_log            = std::stoi(argv[5]);
+    const bool time_kernel       = std::stoi(argv[6]);
+    std::vector<index_t> lengths = std::stoi(argv[7]);
+
+    /**const int N = std::stoi(argv[7]);
+    const int C = std::stoi(argv[8]);
+    const int D = std::stoi(argv[9]);
+    const int H = std::stoi(argv[10]);
+    const int W = std::stoi(argv[11]);**/
+
+    using F32 = float;
+    using F16 = ck::half_t;
+
+    auto profile = [&](auto a_type, auto b_type) {
+        using ADataType = decltype(a_type);
+        using BDataType = decltype(b_type);
+
+        bool pass = ck::profiler::profile_transpose_impl<ADataType, BDataType>(
+            do_verification, init_method, do_log, time_kernel, lengths);
+
+        return pass ? 0 : 1;
+    };
+
+    if(data_type == GemmDataType::F32_F32_F32_F32_F32)
+    {
+        return profile(F32{}, F32{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16_F16_F16)
+    {
+        return profile(F16{}, F16{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_gemm_transpose);

test/transpose/test_transpose_ut_cases.inc

@@ -14,7 +14,6 @@ TYPED_TEST(TestTranspose, Test1)
     this->Run();
 }
 
-
 TYPED_TEST(TestTranpose, Test2)
 {
     std::vector<int> Ms{127, 255, 312, 799, 1573};
@@ -27,4 +26,3 @@ TYPED_TEST(TestTranpose, Test2)
 
     this->Run();
 }
-