Performance Evaluation: A request to Include in JIT comparison

[jserv]

I would like to make a request for the inclusion of a test item based on copyjit for evaluation within the existing jit-comparison project. Currently, the project offers a basic comparison among gccjit, GNU Lightning, libjit, and MIR. To ensure comprehensive coverage of JIT compilers, it would be beneficial to add a test that focuses on copyjit's performance.

copyjit is a notable contender in the JIT landscape, and including it in the evaluation would provide valuable insights into its performance characteristics. This addition would contribute to a more thorough analysis of JIT compilers and enable developers and researchers to make informed decisions about their usage.

[Kimplul]

Good suggestion, I added copyjit along with lightening and two bytecode implementations for reference.

A simple loop probably isn't the best test case, as it seems that most JITs are able to produce near-optimal code (at least when handwritten) for it. I don't really think there's a single test that would cover all aspects of JITs, but a number of smaller tests could be useful. Something like matrix operations and fibonacci sequences come to mind, though implementing them is firmly in the TODO pile for now.

[Kimplul]

I added a recursive fibonacci sequence as a test case to jit-comparison.

Not all libraries have an implementation yet but the ones that do show a larger disparity between each library compared to the loop case, which in this case is probably a good thing.