--snip-- I think it safe to conclude that memcpy (T2) is the fastest all-round solution. Only the inlined custom code can beat it. What's more, to use memcpy only one line in gc.d needs to be changed. Both T3 and T4 would need a change in the compiler. L.