Comment by nneonneo

Now that I'm thinking about it a bit more - I recall that GCC 6.1 added "-fno-plt" for x64 targets, which replaces PLT calls with "call [func@got]" instead, eliminating the double-indirection. Unfortunately, I don't have a GCC 6.1 environment, so I'd be curious to see if someone can run the benchmark with "-fno-plt" and how it performs. A combination of that plus indirect call prediction should basically close the gap with the JIT approach.