[treerex]

Compiler vendors have been doing stuff like this (recognizing specific code patterns from common benchmarks) for 30 years, back in the days when there was real competition amongst C compilers (Borland vs. Zortech, for example.)

[gjm11]

Yup. But the HN discussion of this one seemed to be treating it as an instance of compilers being terribly clever to benefit their actual users, rather than yet another benchmark hack.