Debian Code Search replaced its 7-year-old C TurboPFor library with pure Go using Go 1.26/1.27's experimental simd/archsimd package, achieving 3x speedup via AVX2 vertical layout and a further 2x via AI-suggested AVX-512 positional popcount (GF2P8AFFINEQB + VPOPCNTB), reaching 7 IPC near hardware limits.
The article explains SIMD fundamentals, traces the evolution of x86 instruction sets, compares Intel and AMD implementations, examines the controversy around AVX‑512, and shows how Rust leverages SIMD to accelerate critical paths in high‑frequency quantitative trading, offering concrete code examples and performance tips.
The article examines Go 1.25’s experimental Green Tea garbage collector, explains why the traditional mark‑sweep approach hurts modern CPUs, details the page‑oriented redesign and its AVX‑512 vector acceleration, and shows how these changes can cut GC‑related CPU usage by 10‑40%.
Linus Torvalds lambasts Intel’s AVX‑512 extension as a ‘power virus’, arguing it bloats CPUs, serves only niche benchmarks, and harms everyday performance, while noting Intel’s recent move to limit AVX‑512 to large cores in Alder Lake and recalling his past clashes over Intel’s security patches.
Linus Torvalds publicly denounced Intel’s AVX‑512 instruction set, calling it a ‘power‑virus’ that bloats CPUs for niche benchmark gains, urging Intel to abandon it in favor of simpler, more efficient code that benefits everyday users rather than specialized high‑performance scenarios.
