KernelName,Count,Sum(ns),Mean(ns),Median(ns),Pct
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 512u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,6051953.0,6051953.0,6051953.0,9.170412884681694
"void benchmark_func<int, 256, 8u, 512u>(int, int*) [clone .kd]",1,4525085.0,4525085.0,4525085.0,6.856777934045399
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 256u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,3047976.0,3047976.0,3047976.0,4.618541879390102
"void benchmark_func<double, 256, 8u, 512u>(double, double*) [clone .kd]",1,3041096.0,3041096.0,3041096.0,4.608116742141579
"void benchmark_func<int, 256, 8u, 256u>(int, int*) [clone .kd]",1,2280622.0,2280622.0,2280622.0,3.4557845002908207
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 128u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,1546868.0,1546868.0,1546868.0,2.343940582172697
"void benchmark_func<double, 256, 8u, 256u>(double, double*) [clone .kd]",1,1540948.0,1540948.0,1540948.0,2.334970115237921
"void benchmark_func<__half2, 256, 8u, 512u>(__half2, __half2*) [clone .kd]",1,1539028.0,1539028.0,1539028.0,2.3320607746104267
"void benchmark_func<float, 256, 8u, 512u>(float, float*) [clone .kd]",1,1535188.0,1535188.0,1535188.0,2.3262420933554373
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 96u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,1169910.0,1169910.0,1169910.0,1.772743069537711