Micro-benchmarking .NET Native and RyuJit

[Disclaimer] As both RyuJIT and  .NET Native are improving a lot between their updates/previews, results of benchmarks in this post may no longer be relevant. Benchmarks were ran with the following versions:

• RyuJIT CTP4 
• .NET Native developer preview 2 

While .NET JIT is performing quite well on Windows, it is still behind a fully optimized C++ program, though an efficient compiled program is not only about code generation but also memory management and data locality, the .NET Team has recently introduced two new technologies that might help for the code gen part: The introduction of  .NET Native, an offline .NET compiler (similar to ngen, but using the backend optimizer that is used by the C++ compiler) and the next generation of .NET JIT called "RyuJit". In this post I would like to present the results of some micro-benchmarks that are roughly evaluating some performance benefits of these two new technologies.

First of all, you may have already read about a few benchmarks results available around about RyuJit and .NET Native, here is a non-exhaustive list I have found (if you have more pointers, let me know!):

• A first look at RyuJIT CTP3 and SIMD (SSE2) support for .NET by Frank Niemeyer
• Lies, damn lies and benchmarks by Kevin Frei
• .NET Native Performance by Sasha Goldstein