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]",1,6056598.0,6056598.0,6056598.0,9.176978573582865
"void benchmark_func<int, 256, 8u, 512u>(int, int*) [clone .kd]",1,4526368.0,4526368.0,4526368.0,6.858368700077359
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 256u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,3051019.0,3051019.0,3051019.0,4.62291471063363
"void benchmark_func<double, 256, 8u, 512u>(double, double*) [clone .kd]",1,3042859.0,3042859.0,3042859.0,4.610550649957911
"void benchmark_func<int, 256, 8u, 256u>(int, int*) [clone .kd]",1,2280304.0,2280304.0,2280304.0,3.4551246342014617
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 128u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,1546548.0,1546548.0,1546548.0,2.3433349644499164
"void benchmark_func<double, 256, 8u, 256u>(double, double*) [clone .kd]",1,1540469.0,1540469.0,1540469.0,2.334124042287209
"void benchmark_func<__half2, 256, 8u, 512u>(__half2, __half2*) [clone .kd]",1,1538229.0,1538229.0,1538229.0,2.330729986415443
"void benchmark_func<float, 256, 8u, 512u>(float, float*) [clone .kd]",1,1534229.0,1534229.0,1534229.0,2.324669172358718
"void benchmark_func<HIP_vector_type<float, 2u>, 256, 8u, 96u>(HIP_vector_type<float, 2u>, HIP_vector_type<float, 2u>*) [clone .kd]",1,1167991.0,1167991.0,1167991.0,1.7697440677320215