Cross-compilation

Cross-compilation is the practice of compiling executable code on one computer architecture for execution on a different architecture. A cross-compiler must generate machine code for a target platform — its instruction set, calling conventions, memory model, and system ABI — while running on a host platform with potentially different properties. This architectural separation makes cross-compilation essential for embedded systems, mobile devices, and any target too resource-constrained to host its own toolchain.

The complexity of cross-compilation lies in the abstraction gap between the compiler's model of the target and the target's physical reality. The compiler selects instructions based on a specification, not on measurement. It cannot probe the target's cache hierarchy, branch predictor behavior, or memory latency. The resulting binary is optimized for a Platonic machine that may differ in subtle but performance-critical ways from the actual hardware. This is why profile-guided optimization is rarely effective in cross-compilation scenarios: the training data must come from the target, but obtaining it requires the target to already be capable of running code.