From 813ad5a2ca3f67e188e54c9edf66dffc93283fd7 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Ville=20Pietil=C3=A4?= <>
Date: Wed, 10 Dec 2025 10:34:38 -0500
Subject: [PATCH] Enable running int8 instances.

---
 .../tile_profiler_operation_registry.hpp       |  6 ------
 .../src/tile_profile_grouped_conv_fwd.cpp      | 18 ++++++++++++------
 .../gpu/tile_grouped_conv_fwd_factory.hpp      |  2 +-
 3 files changed, 13 insertions(+), 13 deletions(-)

diff --git a/experimental/ck_tile_profiler/include/tile_profiler_operation_registry.hpp b/experimental/ck_tile_profiler/include/tile_profiler_operation_registry.hpp
index 1f2acb93c4..d22960414c 100644
--- a/experimental/ck_tile_profiler/include/tile_profiler_operation_registry.hpp
+++ b/experimental/ck_tile_profiler/include/tile_profiler_operation_registry.hpp
@@ -56,12 +56,6 @@ class ProfilerOperationRegistry final
     {
         const auto found = entries_.find(name);
 
-        for (const auto& [key, value] : entries_)
-        {
-            // Debug output to trace available operations
-            std::cout << "Registered operation: " << key << " - " << value.description_ << std::endl;
-        }
-
         if(found == end(entries_))
         {
             return std::nullopt;
diff --git a/experimental/ck_tile_profiler/src/tile_profile_grouped_conv_fwd.cpp b/experimental/ck_tile_profiler/src/tile_profile_grouped_conv_fwd.cpp
index 283e6e7fc5..a78d0fd0a9 100644
--- a/experimental/ck_tile_profiler/src/tile_profile_grouped_conv_fwd.cpp
+++ b/experimental/ck_tile_profiler/src/tile_profile_grouped_conv_fwd.cpp
@@ -78,8 +78,8 @@ static void print_helper_msg()
 
 int tile_profile_grouped_conv_fwd(int argc, char* argv[])
 {
-    // 8 for control, 1 for num_dim_spatial
-    if(argc < 9)
+    // 7 for control, 1 for num_dim_spatial
+    if(argc < 8)
     {
         print_helper_msg();
         return 1;
@@ -93,10 +93,12 @@ int tile_profile_grouped_conv_fwd(int argc, char* argv[])
     const bool time_kernel     = std::stoi(argv[7]);
     const int num_dim_spatial  = std::stoi(argv[8]);
 
-    // 9 for control, 1 for num_dim_spatial, 4 for G/N/K/C, and 6 * num_dim_spatial
-    if(argc != 8 + 1 + 4 + 6 * num_dim_spatial + 1)
+    // 7 for control, 1 for num_dim_spatial, 4 for G/N/K/C, and 6 * num_dim_spatial
+    const int expected_num_args = 7 + 1 + 4 + 6 * num_dim_spatial + 1;
+    if(argc != expected_num_args)
     {
-        std::cout << argc << std::endl;
+        std::cout << "Received " << argc << " args"<< std::endl;
+        std::cout << "Expected " << expected_num_args << " args"<< std::endl;
         print_helper_msg();
         return 1;
     }
@@ -170,6 +172,10 @@ int tile_profile_grouped_conv_fwd(int argc, char* argv[])
         {
             return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, BF16{}, BF16{}, BF16{}, BF16{}, BF16{});
         }
+        if (data_type == ConvDataType::I8_I8_I8)
+        {
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, int8_t{}, int8_t{}, int8_t{}, int8_t{}, int8_t{});
+        }
     }
 
     if(num_dim_spatial == 3 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
@@ -187,7 +193,7 @@ int tile_profile_grouped_conv_fwd(int argc, char* argv[])
             return profile(
                 I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, BF16{}, BF16{}, BF16{}, BF16{}, BF16{});
         }
-        else if(data_type == ConvDataType::I8_I8_I8)
+        if(data_type == ConvDataType::I8_I8_I8)
         {
             return profile(
                 I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, int8_t{}, int8_t{}, int8_t{}, int8_t{}, int8_t{});
diff --git a/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_fwd_factory.hpp b/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_fwd_factory.hpp
index 49944f1a1e..5e0e91e2b6 100644
--- a/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_fwd_factory.hpp
+++ b/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_fwd_factory.hpp
@@ -65,7 +65,7 @@ void add_grouped_conv2d_fwd_bf16_instances_6(std::vector<std::unique_ptr<DeviceO
 void add_grouped_conv2d_fwd_f16_instances(std::vector<std::unique_ptr<DeviceOpFwd2DF16>>& instances);
 
 // INT8 instances
-void add_grouped_conv2d_fwd_i8_instances(std::vector<std::unique_ptr<DeviceOpFwd2DINT8>>& instances);
+void add_grouped_conv2d_fwd_int8_instances(std::vector<std::unique_ptr<DeviceOpFwd2DINT8>>& instances);
 
 template <ck_tile::index_t NumDimSpatial,
           typename InLayout,