Hi, Mike, It would be very interesting if you could publish the exact source code used for the benchmarks in the LuaJIT website. I always was curious, for example, if you used FFI native arrays or structs to speed up any benchmarks, or if the code run by the Lua and LuaJIT executables are exactly the same. Well, in the SciMark benchmark, whose source code you published, you used FFI, so I thought you may also have used it in another benchmarks. I know most of the benchmarks are derived from Shootout, but they don't have LuaJIT in their comparison anymore, and your code may be different from the Lua benchmarks listed there. The benchmarks would also be an useful resource for anyone doing modifications to the LuaJIT source code or porting it to a new architecture. Thanks and best regards, Paulo Matias