Even if you're smarter than your compiler it's usually not worth the readability trade-off.
These sorts of optimisations only ever make sense if you're doing something incredibly intensive and you've tested it on real world data. This is rarely the case and the code suffers.
[1]: "Source Code Optimization" => http://www.linux-kongress.org/2009/slides/compiler_survey_fe...
I just tried my age-old adage of doing bit shifts and adds instead of multiplications. Turns out that's slower on an i7!
The cool thing is that llvm-g++ actually recognizes my bad hand-optimization on -O3 and fixes it.
Code at https://gist.github.com/2896137
Or are working on something incredibly small with an incredibly stupid compiler.
For example 99% of the time I want a binary shift and not a /2.In a dynamically typed language 201/2 will give you 100.5 when you'd prefer to keep it integer.
Not necessarily. Python gives 100, PHP gives 100.5, Perl gives 100.5, Ruby gives 100.
Edit: Versions are Python 2.7.3, PHP 5.3.10, Perl 5.14.2 and Ruby 1.8.7, all as shipped with Ubuntu Precise.
Lisp will give you 201/2.