Replace vseq/vand with their immediate-form variants

[heiher]

simd/lines_fwd/134217728
                        time:   [4.3447 ms 4.3474 ms 4.3505 ms]
                        thrpt:  [28.732 GiB/s 28.753 GiB/s 28.771 GiB/s]
                 change:
                        time:   [−1.8221% −1.7185% −1.6176%] (p = 0.00 < 0.05)
                        thrpt:  [+1.6442% +1.7485% +1.8559%]
                        Performance has improved.

[lhecker]

Oh, that's very interesting! I was reading a lot of RISC-V lately (my first time doing so) and this reminds me a lot of it. I hope LoongArch continuous to gain popularity. 🙂

[lhecker]

Apropos, I'm possibly looking to implement this algorithm in the future: http://0x80.pl/notesen/2018-10-18-simd-byte-lookup.html
I could use it for my syntax highlighting PR where I implemented a custom regex engine. Perhaps this could also be of interest for you? I think it may find some great reuse as other regex engines use similar approaches. I'm just not sure if it'll be easily translatable to LoongArch (or even NEON for that matter).

[heiher]

I'm just not sure if it'll be easily translatable to LoongArch (or even NEON for that matter).

It seems that LoongArch SIMD should be able to handle this without much trouble.