Jörn Engel (joern.delete@this.purestorage.com) on July 11, 2020 8:33 pm wrote:

> The main downside is that you now have to modify your algorithms and write code in a language
> that Intel completely messed up. _mm256_permute2x128_si256? _mm256_add_epi16 and _mm256_add_epi32
> instead of using the right type and typing "a += b"? Finding any detail in the programming
> interface that people don't intuitively dislike is hard.
>
> But once you grit your teeth and accept the ugliness (or work around it), explicit parallelism specified
> by the programmer is much more efficient than OOO. I would predict that people will slowly modify
> their code to take advantage of it and the benefit of vector units will increase as they do.

I've done some hobby SSE programming for about fifteen years. At work (game engine programming) we have all the necessary #define SIMD_ENABLED things in codebase, so adding an SSE (or AVX, if it was significantly better) codepath would be pretty straight forward. Been ten years now and I've never run into anything that I've even thought about bothering with. Few cases where we use SIMD are quite traditional graphics stuff (which tends to get moved to GPU as soon as possible) and bit less traditional physics stuff (which is a complete mess and hard to even measure any improvement).

What sort of things are you seeing in regular code that you'd get some normal programmers casually converting to SIMD?

-JLarja