XOR swap (Stephan Brumme, bits)

stephan-brumme‘s page on swapping two values without a temporary, from his benchmarked bits.stephan-brumme.com collection (each trick shipped with assembler output, timings, and source).

The trick

Three XORs exchange two integers using the identity a ^ b ^ a = b:

a ^= b;
b ^= a;
a ^= b;

The caveats (the reason it’s a cautionary tale, not advice)

• Aliasing hazard: must not be applied to the same storage twice — swapXor(x, x) zeroes the value (each ^= against itself yields 0). The temp-variable swap has no such trap.
• Not faster on modern CPUs: Brumme benchmarks it as slower than the obvious temp-variable swap across Intel processors — registers and store-to-load forwarding make the temp version cheaper, and the three XORs carry a serial dependency chain. The trick’s old appeal (saving a register) no longer pays.

Why it’s here

The cleanest single illustration of this spoke’s central lesson (branchless-programming): a famous bit trick that is elegant, correct-with-caveats, and obsolete as an optimization. Belongs with the other Brumme/bit-twiddling-hacks entries as a technique to understand rather than deploy.

Tier

T2 — an author-maintained technical reference with disassembly + benchmarks (neutral, no vendor interest), though single-author and self-published. Benchmark figures are dated/CPU-specific snapshots.

Related

branchless-programming · bit-manipulation · stephan-brumme · branchless-abs · bit-twiddling-hacks