About this role
Falcon is the optimizing just in time (JIT) compiler in Azul Platform Prime. It compiles hot Java bytecode into highly optimized machine code and delivers Java performance that standard JVMs do not reach. Falcon is built on LLVM.
We are looking for a Compiler Performance Engineer to make Falcon generate even faster code. This is not a compiler front end role and not a compiler built from scratch: the compiler exists, runs in production at large enterprises, and your job is to find where the generated code is slower than it should be and fix it in the optimizer and code generator.
You will join a distributed team of compiler experts and work on real workloads with direct, measurable impact on customer performance.
- Analyze performance of Java benchmarks and customer workloads, find gaps in generated code and trace them to their root cause in the compiler
- Implement and improve optimizations in Falcon's middle end and back end: inlining, loop optimizations, vectorization, redundancy elimination, speculative optimizations and their interaction with the runtime
- Improve code generation for x86_64 and AArch64
- Write tests and benchmarks that prove the improvement and protect it from regressions
- Contribute changes upstream to LLVM where it makes sense
- 2+ years of professional experience in compilers, JIT compilers or low level performance engineering
- Strong C++ design and coding skills
- Solid understanding of compiler optimization theory: SSA, data flow analysis, loop transformations
- Ability to read assembly and reason about why one instruction sequence is faster than another
- Experience with LLVM IR and LLVM optimization passes, or with another optimizing compiler such as HotSpot C2, Graal, GCC or V8
- Experience with profilers such as perf, VTune or async profiler
- Knowledge of x86_64 and/or AArch64 microarchitecture
- Experience with Java, Scala or Kotlin, or with JVM internals
- MSc or PhD in Computer Science with a compiler or performance focus
Tech stack
JavaScalaKotlinC++