This is the n-body problem from The Computer Language Benchmarks Game site, written in x86-64 assembly.
It's not ready, I only wrote it as a learning experience.
It's ~1.9x slower than the optimized C version, optimizations still missing include:
- Using SIMD instructions. (
movapd,addpd,subpd,rsqrtps, etc) - Calculating
dx,dy, anddz, magnitude, velocity, and final position each in their own loop to improve cache coherence and to calculate multiple bodies at the same time. - Use
rsqrtpswith two Newton's method steps.
nasm -felf64 nbody.asm && gcc -no-pie nbody.o -o nbody
./nbody <number here>
(If you don't pass it a number it will segfault :))