C · 6 · Advanced / Job-Ready34 / 35 · 97%

Profiling & Performance

Measure first. The bottleneck is almost never where you guessed.

Examples: perf, gprof, cache misses, -O2

shortcuts: ← prev · → next · M mark

Tools

Sampling profilers show where the time actually goes.

Syntax

syntax

1
2
3

perf stat ./app                 # cycles, IPC, cache misses
perf record -g ./app && perf report
gcc -pg app.c -o app && ./app && gprof ./app

Cache Locality Wins

Row-major traversal is often several times faster than column-major.

Example

example

// fast: sequential memory
for (int i = 0; i < N; i++)
    for (int j = 0; j < N; j++) sum += m[i][j];

// slow: strides the cache on every step
for (int j = 0; j < N; j++)
    for (int i = 0; i < N; i++) sum += m[i][j];

TIP

Same complexity, same instruction count — the only difference is cache behaviour.

Compiler Flags

The optimizer is your cheapest speedup.

Complexity

-O0	no opt	debug builds only
-O2	default release
-O3	more inlining/vectorizing	measure, can be slower
-march=native	uses this CPU's ISA	not portable
-flto	cross-file inlining

Method

A repeatable loop beats intuition.

Interview question

How do you approach optimizing a slow C program?