Writing Efficient Programs
Also has a great bunch of "war stories" about performance tuning in real-life, including one in which people, IIRC, improve the performance of quicksort on a supercomputer by 1 million times or some such, by working on tuning as well as architecture and algorithms at several levels of the stack, from the hardware on upwards.
by:
https://en.wikipedia.org/wiki/Jon_Bentley_(computer_scientis...
Edited to change Wikipedia URL to more specific one.