Benchmarks
The same Haxe source on every target
Each kernel runs unchanged on Rayzor, hxcpp, HashLink, HashLink/C and the JVM. The set covers
recursion, floating-point loops, object graphs and allocation, plus the Computer Language Benchmarks Game
kernels: permutations, matrix-vector products, sequence generation, hashing, string reversal, bignum
arithmetic and regular expressions. CI measures every push to main.
lower is better
Loading benchmark results…
Bars show the mean of the measured iterations after warm-up; hover over a time for its spread, iteration
count and compile time. Execution times the program alone.
Compile + run adds what it takes to get there: Rayzor parses, types and JIT-compiles in-process
in tens of milliseconds, while hxcpp, HashLink/C and the JVM build ahead of time (tagged
AOT) and their build is added to the bar.
Rayzor is what
rayzor run does: start on Cranelift and promote hot code to LLVM. All tiers also shows the
Cranelift-only and LLVM-only modes and the precompiled bundles (RZB).
Each kernel runs as its own CI job, so different rows may come from different machines: compare targets
within a row. Kernel sources live in compiler/benchmarks.