Skip to content

Comment on Fast inverse square rootparent

Comments

Could duplicate the x264_log2_lut table, then you wouldn't need the 'and eax, 0x7f'.... 6 instructions.

Yeah, though at that point I'd be start to worry about whether that's a fair L1 cache tradeoff, given that it only saves a single (integer!) instruction and requires 512 bytes of memory.

Sure, always a trade off when you get to that sort of point, and not much payoff.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.