Enhance printing functionality (#1751)

* Added object print with all template parameters

* fix clang format

---------
Co-authored-by: ravil-mobile <ravil.aviva.com@gmail.com>
Co-authored-by: illsilin <Illia.Silin@amd.com>

include/ck/tensor_operation/gpu/device/device_base.hpp

@@ -5,6 +5,8 @@
 
 #include <string>
 #include <sstream>
+#include <regex>
+#include <optional>
 
 #include "ck/stream_config.hpp"
 
@@ -12,6 +14,34 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
+#define GET_OBJECT_NAME_IMLP                                                  \
+    std::optional<std::string> GetObjectName() const override                 \
+    {                                                                         \
+        std::string str = __PRETTY_FUNCTION__;                                \
+        static std::regex obj_name_expr{"<std::string> (.*)::GetObjectName"}; \
+        std::smatch match;                                                    \
+        if(!std::regex_search(str, match, obj_name_expr))                     \
+        {                                                                     \
+            return str;                                                       \
+        }                                                                     \
+        return std::string(match[1]) + ';';                                   \
+    }
+
+#define GET_TEMPLATE_INFO_IMPL                                  \
+    std::optional<std::string> GetTemplateInfo() const override \
+    {                                                           \
+        std::string str = __PRETTY_FUNCTION__;                  \
+        static std::regex template_expr{"\\[(.*)\\]"};          \
+        std::smatch match;                                      \
+        if(!std::regex_search(str, match, template_expr))       \
+        {                                                       \
+            return std::nullopt;                                \
+        }                                                       \
+        return std::string(match[1]);                           \
+    }
+
+#define REGISTER_EXTRA_PRINTING_METHODS GET_OBJECT_NAME_IMLP GET_TEMPLATE_INFO_IMPL
+
 struct BaseArgument
 {
     BaseArgument()                    = default;
@@ -48,6 +78,10 @@ struct BaseOperator
 
     virtual std::string GetTypeIdName() const { return typeid(*this).name(); }
 
+    virtual std::optional<std::string> GetObjectName() const { return std::nullopt; }
+
+    virtual std::optional<std::string> GetTemplateInfo() const { return std::nullopt; }
+
     virtual std::string GetTypeIdHashCode() const
     {
         std::ostringstream oss;

include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle_v3.hpp

@@ -729,6 +729,7 @@ struct DeviceGemm_Xdl_CShuffleV3 : public DeviceGemmV2<ALayout,
 
         return str.str();
     }
+    REGISTER_EXTRA_PRINTING_METHODS
 };
 
 } // namespace device

profiler/include/profiler/profile_gemm_universal_impl.hpp

@@ -144,6 +144,7 @@ bool profile_gemm_universal_impl(int do_verification,
     }
 
     std::string best_op_name;
+    std::optional<std::string> best_op_object_name;
     float best_ave_time   = 0;
     float best_tflops     = 0;
     float best_gb_per_sec = 0;
@@ -225,7 +226,8 @@ bool profile_gemm_universal_impl(int do_verification,
                     }
                 }
 
-                std::string op_name = op_ptr->GetTypeString();
+                std::string op_name                    = op_ptr->GetTypeString();
+                std::optional<std::string> op_obj_name = op_ptr->GetObjectName();
 
                 float ave_time = invoker_ptr->Run(argument_ptr.get(),
                                                   StreamConfig{nullptr,
@@ -251,11 +253,12 @@ bool profile_gemm_universal_impl(int do_verification,
 
                 if(tflops > best_tflops && ave_time > 1e-10)
                 {
-                    best_op_name    = op_name;
-                    best_tflops     = tflops;
-                    best_ave_time   = ave_time;
-                    best_gb_per_sec = gb_per_sec;
-                    best_kbatch     = kbatch_curr;
+                    best_op_name        = op_name;
+                    best_op_object_name = op_obj_name;
+                    best_tflops         = tflops;
+                    best_ave_time       = ave_time;
+                    best_gb_per_sec     = gb_per_sec;
+                    best_kbatch         = kbatch_curr;
                 }
             }
             else
@@ -306,6 +309,9 @@ bool profile_gemm_universal_impl(int do_verification,
               << " : " << best_ave_time << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec
               << " GB/s, " << best_op_name << std::endl;
 
+    if(best_op_object_name)
+        std::cout << best_op_object_name.value() << std::endl;
+
     return pass;
 }