From 3aec8b5493c0ca6c272dacbc305f761aabf2373e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Ville=20Pietil=C3=A4?= <ville.pietila@amd.com>
Date: Mon, 20 Oct 2025 08:04:59 +0000
Subject: [PATCH] Improve profiler output.

---
 ...e_profile_grouped_conv_bwd_weight_impl.hpp | 36 ++++++++++++++-----
 .../profile_grouped_conv_bwd_weight_impl.hpp  | 30 ++++++++++++++--
 2 files changed, 55 insertions(+), 11 deletions(-)

diff --git a/profiler/ck_tile/include/tile_profile_grouped_conv_bwd_weight_impl.hpp b/profiler/ck_tile/include/tile_profile_grouped_conv_bwd_weight_impl.hpp
index a4316bfa3d..7906007369 100644
--- a/profiler/ck_tile/include/tile_profile_grouped_conv_bwd_weight_impl.hpp
+++ b/profiler/ck_tile/include/tile_profile_grouped_conv_bwd_weight_impl.hpp
@@ -8,6 +8,7 @@
 #include <iostream>
 #include <iterator>
 #include <typeinfo>
+#include <sstream>
 
 #include "ck_tile/host.hpp"
 #include "ck_tile/host/convolution_parameter.hpp"
@@ -257,14 +258,33 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
         }
     }
 
-    std::cout << "\n********************************" 
-              << "\nBest configuration parameters:" 
-              << "\n********************************" 
-              << "\nname: " << best_op_name
-              << "\navg_time: " << best_avg_time << "\ntflops: " << best_tflops
-              << "\nGB/s: " << best_gb_per_sec 
-              << "\nSplitK: " << best_split_k
-              << std::endl;
+    std::stringstream ss;
+    ss << "\n********************************" 
+       << "\nCK Tile best configuration parameters:" 
+       << "\n********************************" 
+       << "\nname: " << best_op_name
+       << "\navg_time: " << best_avg_time << "\ntflops: " << best_tflops
+       << "\nGB/s: " << best_gb_per_sec 
+       << "\nSplitK: " << best_split_k
+       << std::endl;
+
+    std::cout << ss.str();
+
+    const char* log_file = std::getenv("CK_TILE_PROFILER_LOG_FILE");
+    if(log_file != nullptr)
+    {
+        std::ofstream out(log_file, std::ios::app);
+        if(out.is_open())
+        {
+            std::stringstream out_ss;
+            out_ss << "CK Tile best configuration:" << std::endl
+                   << "name: " << best_op_name << std::endl
+                   << "avg_time: " << best_avg_time << std::endl
+                   << "SplitK: " << best_split_k << std::endl;
+            out << out_ss.str();
+            out.close();
+        }
+    }
 
     if(instance_index != -1)
     {
diff --git a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
index 6654275fd0..4dc0f7115e 100644
--- a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
@@ -8,6 +8,7 @@
 #include <iostream>
 #include <iterator>
 #include <typeinfo>
+#include <sstream>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
@@ -330,9 +331,32 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
         }
     }
 
-    std::cout << "Best configuration parameters:" << "\nname: " << best_op_name
-              << "\navg_time: " << best_avg_time << "\ntflops: " << best_tflops
-              << "\nGB/s: " << best_gb_per_sec << ", SplitK " << best_split_k << std::endl;
+    std::stringstream ss;
+    ss << "\n********************************" 
+       << "\nCK best configuration parameters:" 
+       << "\n********************************" 
+       << "\nname: " << best_op_name
+       << "\navg_time: " << best_avg_time << "\ntflops: " << best_tflops
+       << "\nGB/s: " << best_gb_per_sec 
+       << "\nSplitK: " << best_split_k
+       << std::endl;
+    std::cout << ss.str();
+
+    const char* log_file = std::getenv("CK_PROFILER_LOG_FILE");
+    if(log_file != nullptr)
+    {
+        std::ofstream out(log_file, std::ios::app);
+        if(out.is_open())
+        {
+            std::stringstream out_ss;
+            out_ss << "CK best configuration:" << std::endl
+                   << "name: " << best_op_name << std::endl
+                   << "avg_time: " << best_avg_time << std::endl
+                   << "SplitK: " << best_split_k << std::endl;
+            out << out_ss.str();
+            out.close();
+        }
+    }
 
     if(instance_index != -1)
     {