review changes: code cleanup

Author: arai713

projects/composablekernel/test/ck_tile/CMakeLists.txt

@@ -38,6 +38,8 @@ add_subdirectory(core)
 add_subdirectory(epilogue)
 add_subdirectory(atomic_add_op)
 add_subdirectory(fmha)
+# TODO: The Universal GEMM tile engine test will be either removed
+# or moved to the appropriate location in future work.
 # add_subdirectory(gemm_tile_engine)
 add_subdirectory(pooling)
 add_subdirectory(grouped_conv)

projects/composablekernel/tile_engine/ops/common/utils.hpp

@@ -9,6 +9,7 @@
 #include <sstream>
 #include <vector>
 #include <string>
+#include <cstdlib>
 
 #include "ck_tile/core.hpp"
 #include "ck_tile/host.hpp"
@@ -106,7 +107,9 @@ struct Setting
 
 inline std::string get_rocm_version()
 {
-    std::ifstream version_file("/opt/rocm/.info/version");
+    const char* path      = std::getenv("ROCM_PATH");
+    std::string file_name = std::string(path) + "/.info/version";
+    std::ifstream version_file(file_name);
     if(version_file.is_open())
     {
         std::string version;

projects/composablekernel/tile_engine/ops/gemm/gemm_multi_d/gemm_multi_d_benchmark.hpp

@@ -14,7 +14,7 @@
 #include "gemm/gemm_benchmark.hpp"
 
 #pragma clang diagnostic push
-#pragma clang diagnostic ignored "-Wlifetime-safety-intra-tu-seggestions"
+#pragma clang diagnostic ignored "-Wlifetime-safety-intra-tu-suggestions"
 
 // Data types and Layouts are defined by the generated kernel headers
 // No hardcoded type definitions here to avoid conflicts

projects/composablekernel/tile_engine/ops/gemm/gemm_multi_d/gemm_multi_d_benchmark.py

@@ -1,3 +1,4 @@
+#!/usr/bin/env python3
 # Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
 # SPDX-License-Identifier: MIT
 
@@ -9,7 +10,7 @@
 
 
 def _import_gemm_benchmark():
-    """Import validation utilities from commons directory."""
+    """Import gemm benchmark from parent directory."""
     current_dir = os.path.dirname(os.path.abspath(__file__))
     parent_dir = os.path.dirname(current_dir)
 

projects/composablekernel/tile_engine/ops/gemm/gemm_multi_d/gemm_multi_d_profiler.hpp

@@ -149,10 +149,15 @@ class GemmMultiDProfiler : public GemmProfiler<GemmMultiDProfiler,
 
         for(auto& callable : callables)
         {
-            auto kernel_run_result =
-                callable(gemm_multi_d_args,
-                         ck_tile::stream_config{
-                             nullptr, true, setting_.log_, setting_.n_warmup_, setting_.n_repeat_});
+            auto kernel_run_result = callable(gemm_multi_d_args,
+                                              ck_tile::stream_config{nullptr,
+                                                                     true,
+                                                                     setting_.log_,
+                                                                     setting_.n_warmup_,
+                                                                     setting_.n_repeat_,
+                                                                     setting_.is_gpu_timer_,
+                                                                     setting_.flush_cache_,
+                                                                     setting_.rotating_count_});
             process_result(gemm_multi_d_problem,
                            c_m_n_dev_buf,
                            c_m_n_host_result,

projects/composablekernel/tile_engine/ops/gemm/gemm_preshuffle/gemm_preshuffle_benchmark.py

@@ -1,3 +1,4 @@
+#!/usr/bin/env python3
 # Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
 # SPDX-License-Identifier: MIT
 
@@ -9,7 +10,7 @@
 
 
 def _import_gemm_benchmark():
-    """Import validation utilities from commons directory."""
+    """Import gemm benchmark from parent directory."""
     current_dir = os.path.dirname(os.path.abspath(__file__))
     parent_dir = os.path.dirname(current_dir)
 

projects/composablekernel/tile_engine/ops/gemm/gemm_preshuffle/gemm_preshuffle_profiler.hpp

@@ -89,7 +89,7 @@ class GemmPreshuffleProfiler
                                 gemm_problem.stride_c_);
         }
 
-        // Kerenl Execution
+        // Kernel Execution
 
         a_m_k_dev_buf.ToDevice(a_m_k.data());
         c_m_n_dev_buf.SetZero();

projects/composablekernel/tile_engine/ops/gemm/gemm_profiler.hpp

@@ -68,7 +68,7 @@ class GemmProfiler
             ck_tile::static_for<0, DDataType::size(), 1>{}([&](auto i) {
                 using DType = ck_tile::remove_cvref_t<std::tuple_element_t<i, DDataType>>;
                 num_byte += sizeof(DType) * gemm_problem.m_ * gemm_problem.n_;
-                flop += sizeof(DType) * gemm_problem.m_ * gemm_problem.n_;
+                flop += gemm_problem.m_ * gemm_problem.n_;
             });
         }
 

projects/composablekernel/tile_engine/ops/gemm/gemm_universal/gemm_universal_benchmark.py

@@ -10,7 +10,7 @@
 
 
 def _import_gemm_benchmark():
-    """Import validation utilities from commons directory."""
+    """Import gemm benchmark from parent directory."""
     current_dir = os.path.dirname(os.path.abspath(__file__))
     parent_dir = os.path.dirname(current_dir)