interpreters

Stack-based and register-based VMs differ in how they handle operands. Stack-based VMs push and pop values to an execution stack, while register-based VMs read and write to specific registers, often reducing instruction count but increasing per-instruction complexity.

NaN boxing works by storing pointers and integers inside the unused bits of NaN floating-point values. This technique allows a language runtime to represent multiple data types efficiently in a single machine word without requiring separate type tagging memory.

The best dispatch strategy for an interpreter loop often involves computed goto or threaded code rather than a simple switch statement. Computed goto reduces branch prediction overhead by jumping directly to the next instruction handler, significantly optimizing interpreter performance.

You can build a basic JIT compiler using mmap in C++ by allocating executable memory pages. This allows your language runtime to write generated machine code into memory dynamically and execute it directly, bypassing the standard interpreter loop for hot code paths.