add cold warmpup

example/35_splitK_gemm/run_splitK_gemm_example.inc

@@ -157,7 +157,7 @@ bool run_splitK_gemm(const ProblemSize& problem_size, const ExecutionConfig& con
 
     if(config.time_kernel)
     {
-        float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+        float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel, 0, 300, 500});
 
         std::size_t flop = std::size_t(2) * M * N * K;
         std::size_t num_btype =

include/ck/host_utility/kernel_launch.hpp

@@ -33,10 +33,13 @@ float launch_and_time_kernel(const StreamConfig& stream_config,
         printf("Warm up 1 time\n");
 #endif
         // warm up
+        for(int i = 0; i < stream_config.cold_niters_; ++i)
+        {
             kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
             hip_check_error(hipGetLastError());
+        }
 
-        const int nrepeat = 10;
+        const int nrepeat = stream_config.nrepeat_;
 #if DEBUG_LOG
         printf("Start running %d times...\n", nrepeat);
 #endif

include/ck/stream_config.hpp

@@ -11,4 +11,6 @@ struct StreamConfig
     hipStream_t stream_id_ = nullptr;
     bool time_kernel_      = false;
     int log_level_         = 0;
+    int cold_niters_       = 1;
+    int nrepeat_           = 10;
 };