The random range generation approach from lemire.me/blog/2016/06/3 will be great for further optimizing github.com/AndroidHardeni.
It'll be a nice follow-up to using libdivide for replacing the division instructions used to find the slab and slot indexes: twitter.com/DanielMicay/st.
Quote Tweet
It completely solves a huge performance problem I would have had to deal with by hand and only added 16 bytes to the binary size.
Before:
478.092526 msec
2,000,329,069 cycles
3,525,042,263 instructions
After:
322.796288 msec
1,349,630,342 cycles
3,625,044,439 instructions
Show this thread