(pehringer.info)

302 points pehringer | 4 comments | 18 Oct 24 14:36 UTC | HN request time: 0.446s | source

A quick dive into the Plan9 assembly I picked up while developing my SIMD package for Go, and how it led to a 450% performance boost in calculations.

1. pphysch ◴[18 Oct 24 15:14 UTC] No.41880175[source]▶

This is neat. Why do you think this is not already part of the standard library?

2. pehringer ◴[18 Oct 24 15:33 UTC] No.41880358[source]▶

I think the Go std lib is focused on portability over raw performance

replies(1): >>41887988 #

3. twoodfin ◴[18 Oct 24 16:43 UTC] No.41881109[source]▶

I think the more interesting architecture / priorities question is why the Go compiler can't yet(?) auto-vectorize?

4. pjmlp ◴[19 Oct 24 14:34 UTC] No.41887988[source]▶

And ironically loses out to Java and .NET in raw performance, because language author's bias.

In this case specifically, SIMD is available, via compiler supported auto-vectorization, and intrisics, no need to go down into Assembly.

Show HN: Go Plan9 Memo