KernelName,Count,Sum(ns),Mean(ns),Median(ns),Pct
"void benchmark_func<int, 256, 8u, 512u>(int, int*) [clone .kd]",1,3358711.0,3358711.0,3358711.0,7.846568285268675
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 512u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,1724475.0,1724475.0,1724475.0,4.028691615247248
"void benchmark_func<double, 256, 8u, 512u>(double, double*) [clone .kd]",1,1715675.0,1715675.0,1715675.0,4.0081331924146895
"void benchmark_func<int, 256, 8u, 256u>(int, int*) [clone .kd]",1,1697915.0,1697915.0,1697915.0,3.966642557243526
"void benchmark_func<__half2, 256, 8u, 512u>(__half2, __half2*) [clone .kd]",1,1674395.0,1674395.0,1674395.0,3.911695499854689
"void benchmark_func<float, 256, 8u, 512u>(float, float*) [clone .kd]",1,1665276.0,1665276.0,1665276.0,3.890391834194451
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 256u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,884957.0,884957.0,884957.0,2.067422749390022
"void benchmark_func<double, 256, 8u, 256u>(double, double*) [clone .kd]",1,879838.0,879838.0,879838.0,2.055463821380947
"void benchmark_func<int, 256, 8u, 128u>(int, int*) [clone .kd]",1,869117.0,869117.0,869117.0,2.0304175882914177
"void benchmark_func<__half2, 256, 8u, 256u>(__half2, __half2*) [clone .kd]",1,858557.0,858557.0,858557.0,2.005747480892348