Clean up.

include/ck/utility/work_scheduling.hpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -75,12 +75,11 @@ class StridedReductionTileLoop
     ///
     /// @return     The workgroup flag index.
     ///
-    __device__ uint32_t GetWorkgroupFlagIdx([[maybe_unused]] index_t k_tiles,
+    __device__ uint32_t GetWorkgroupFlagIdx(index_t k_tiles,
                                             index_t output_tile_idx,
                                             index_t output_tile_idx_offset) const
     {
         return (output_tile_idx + output_tile_idx_offset) % GetFlagCount(k_tiles);
-        // return output_tile_idx + output_tile_idx_offset;
     }
 
     ///

profiler/include/profiler/profile_grouped_gemm_multiple_d_splitk_impl.hpp

@@ -307,12 +307,12 @@ bool profile_ggemm_multid_splitk(int do_verification,
                               << (instance_pass ? "SUCCEED" : "FAILED") << std::endl;
 
                     pass = pass && instance_pass;
-                    std::cout << ">>>>>CPU verification end!" << std::endl;
+                    // std::cout << ">>>>>CPU verification end!" << std::endl;
                 }
 
                 if(time_kernel)
                 {
-                    std::cout << ">>>>>GPU time profiling start!" << std::endl;
+                    // std::cout << ">>>>>GPU time profiling start!" << std::endl;
                     float avg_time = invoker_ptr->Run(
                         argument_ptr.get(),
                         StreamConfig{nullptr, time_kernel, 0, warmup_iter, kernel_iter});

profiler/src/profile_grouped_gemm_multiple_d_splitk.cpp

@@ -91,9 +91,9 @@ int profile_grouped_gemm_multiple_d_splitk(int argc, char* argv[])
     const auto StrideAs   = argToIntArray(argv[11]);
     const auto StrideBs   = argToIntArray(argv[12]);
     const auto StrideCs   = argToIntArray(argv[13]);
-    const int kbatch      = argc == 15 ? std::stoi(argv[14]) : 1;
-    const int warmup_iter = argc == 16 ? std::stoi(argv[15]) : 1;
-    const int kernel_iter = argc == 17 ? std::stoi(argv[16]) : 10;
+    const int kbatch      = argc >= 15 ? std::stoi(argv[14]) : 1;
+    const int warmup_iter = argc >= 16 ? std::stoi(argv[15]) : 1;
+    const int kernel_iter = argc >= 17 ? std::stoi(argv[16]) : 10;
 #ifdef CK_ENABLE_FP16
     if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {