6.2. Avoid Unnecessary Memory Accesses

6.2. Avoid Unnecessary Memory Accesses
	Chapter 6. Optimization Workflow

Generally speaking, huge performance improvements come from algorithm changes.

Use an ordinary execution profiler (such as gprof or callgrind) to find hot-spots in the code. That will estimate the amount of time spent in each part and direct your attention to highly used code sections. Evaluate issues like:

Are the right compiler flags used? Is the best compiler used? Some compilers are better than others on optimizing the code.
Is this the best algorithm to perform the job?
Is the program doing unnecessary work, such as unnecessarily repeating an expensive operation?
Can data that is expensive to calculate be cached? For instance, use "dirty" flags to avoid scanning large structures for changes.
Can loops be unrolled?
Can vector instructions be used?
Do loops contain conditional branching? Sometimes these parts can be moved out of the loop.
Is the compiler forced to be conservative about register allocation (due to potential aliasing)? Consider helping the compiler to optimize by introducing temporary variables.
Does the program contain unnecessary copying of structures? This often occurs with programming paradigms that advocate layering, encapsulation and data hiding. Consider relaxing encapsulation rules where it offers performance improvements.

	Tip
	Use optimizing compilers and turn on suitable optimization flags.

	Tip
	Avoid unnecessary work by caching data

	Tip
	Use the correct algorithms

	Tip
	Consider favoring algorithms that touch data sequentially

	Tip
	Help the compiler by rearranging loops and introducing temporary variables

	Tip
	Judiciously break encapsulation to avoid copying data


6.1. Initial State: Correct, Measurable Program, Good Test Case		6.3. Optimize Data Layout