On Binary Translation and Its Consequences
Summary
The post analyzes binary translation of x86-64 to aarch64 on Windows via Prism, assessing performance penalties with Geekbench 7 across CPU architectures (Qualcomm, Neoverse). It discusses translation overhead, instruction expansion, register mapping, and how high-performance cores mitigate but do not eliminate the overhead, concluding that optimal architecture is crucial for viable ARM on Windows performance.