KernelName,Count,Sum(ns),Mean(ns),Median(ns),Pct
"void benchmark_func<int, 256, 8u, 512u>(int, int*) [clone .kd]",1,3357122.0,3357122.0,3357122.0,7.843415692980662
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 512u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,1724161.0,1724161.0,1724161.0,4.028245456860141
"void benchmark_func<double, 256, 8u, 512u>(double, double*) [clone .kd]",1,1715521.0,1715521.0,1715521.0,4.008059383316388
"void benchmark_func<int, 256, 8u, 256u>(int, int*) [clone .kd]",1,1698081.0,1698081.0,1698081.0,3.967313420052145
"void benchmark_func<__half2, 256, 8u, 512u>(__half2, __half2*) [clone .kd]",1,1672801.0,1672801.0,1672801.0,3.908250464127829
"void benchmark_func<float, 256, 8u, 512u>(float, float*) [clone .kd]",1,1664321.0,1664321.0,1664321.0,3.8884382067608114
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 256u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,884640.0,884640.0,884640.0,2.066829641174319
"void benchmark_func<double, 256, 8u, 256u>(double, double*) [clone .kd]",1,879841.0,879841.0,879841.0,2.055617492223339
"void benchmark_func<int, 256, 8u, 128u>(int, int*) [clone .kd]",1,869600.0,869600.0,869600.0,2.0316909205611187
"void benchmark_func<__half2, 256, 8u, 256u>(__half2, __half2*) [clone .kd]",1,859040.0,859040.0,859040.0,2.0070190528965317