LLVM

BayLibre develops and extends LLVM toolchains to support modern CPUs, SoCs, and accelerators across embedded, HPC, and heterogeneous systems. Our work spans frontend integration, IR-level optimizations, backend and code-generation improvements, and architecture-specific enablement for targets such as RISC-V, ARM, and custom accelerators. We focus on performance-critical areas including vectorization, instruction selection, scheduling, memory optimizations, and cross-compilation workflows, while maintaining strict alignment with upstream LLVM development.

Beyond the compiler itself, BayLibre operates LLVM as part of a complete system toolchain. We integrate LLVM with runtimes, debuggers, linkers, operating systems, and BSPs, and address real-world constraints such as determinism, reproducibility, long-term support, and multi-toolchain coexistence. Our engineers regularly work on MLIR-based flows, OpenMP and offload pipelines, and accelerator toolchains, enabling customers to move from hardware bring-up to optimized production binaries with a single, coherent compiler stack.