Benchmarking is done against "hand-rolled" switch-case statements. The benchmark problem is inspired by mpark variant's execute.mpark.cpp
number 4 8 16 32 64 128
implementation
integral_switch_visit_nothrow 2168.70 3110.49 3149.30 3170.34 2999.82 2662.73
switch_case_visit_nothrow 2214.68 3185.18 3127.04 3150.91 3011.64 2629.04