Hacker News Reader

I wonder why writing SIMD in high-level languages hasn't been figured out yet for CPUs (it has been the norm for GPUs for since forever). Auto-vectorization universally sucks, so do OpenMP directives.

There was Ispc, which was a separate C-like programming language just for SIMD, but I don't understand why can't regular compilers generated high-quality vectorized code.

View on HN

Why do you say that? I would say SIMD is pretty well figured out in well-written code, e.g. small, tight loops over vectors. Unrolling and vectorizing a loop is not that hard and happens constantly on all our phones for signal processing, for example.

.NET (C#) is getting there with Vector<T>.

That's just syntactic sugar (and a bit of architecture independence) over intrinsics. You can get the same in C++ just with wrapping intrinsics in classes, and a few ifdefs.

Reply on news.ycombinator.com