clang-17 1brc.c -std=c17 -march=native -mtune=native -Ofast -o 1brc
AVX2 is required. GCC also works, but its binary is ~15% slower.
./1brc measurments.txt n
n is the number of threads to use
Work is spread evenly across threads with no rebalancing. If using hypethreading, n must equal the number of available logical cores for best performance. Use taskset to limit the available logical cores.
Fastest known solution: 0.577s (8 core Zen2); C with heavy AVX2