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]",1,6052433.0,6052433.0,6052433.0,9.171253897560876
"void benchmark_func<int, 256, 8u, 512u>(int, int*) [clone .kd]",1,4524925.0,4524925.0,4524925.0,6.856620476826534
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 256u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,3048937.0,3048937.0,3048937.0,4.620055330586488
"void benchmark_func<double, 256, 8u, 512u>(double, double*) [clone .kd]",1,3041097.0,3041097.0,3041097.0,4.608175375772138
"void benchmark_func<int, 256, 8u, 256u>(int, int*) [clone .kd]",1,2281583.0,2281583.0,2281583.0,3.457283538927013
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 128u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,1547186.0,1547186.0,1547186.0,2.344451501198216
"void benchmark_func<double, 256, 8u, 256u>(double, double*) [clone .kd]",1,1539668.0,1539668.0,1539668.0,2.333059473099456
"void benchmark_func<__half2, 256, 8u, 512u>(__half2, __half2*) [clone .kd]",1,1539028.0,1539028.0,1539028.0,2.3320896808697134
"void benchmark_func<float, 256, 8u, 512u>(float, float*) [clone .kd]",1,1533908.0,1533908.0,1533908.0,2.3243313430317705
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 96u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,1169590.0,1169590.0,1169590.0,1.772280146851395