Stream: git-wasmtime

Topic: wasmtime / issue #5885 fuzzgen: Initial SIMD support

Wasmtime GitHub notifications bot (Feb 28 2023 at 11:15):

afonso360 commented on issue #5885:

I'm not entirely sure I like writing vector constants as large hexadecimal numbers, but it does seem to be the most expedient option here and is apparently rarely used anyway. So let's go with it and maybe we can figure something better out later.

Sounds good. I'm happy to iterate on this syntax as well. From what I've read of the parser we support a few different forms:

const0 = i32x4 0x0123456...
const1 = i64x2 [0 1]
const2 = [0x00 0x01 0x02]

This is pre-this-pr, I've only changed the const2 syntax into big hex. Ideally we could print the const1 format, but that one does require us knowing the type of the const data. Which I don't think we have when printing the CLIF back out.

I'm eager to get the codegen-meta all-opcodes fuzzing stuff going, but I'd still be happy to approve PRs adding more ops to the existing table. I'd like to have as much coverage from fuzzing as possible

Yeah, that makes sense! I'm going to review whats already supported in the interpreter and add those ops to the fuzzer.

and I expect it's going to be tricky to get the code-generated version working, so I'm assuming we won't finish that soon.

Last I looked into it, I ran into some kind of blocker, but I don't actually remember what it was. I'll need to go look into it again.

Last updated: Dec 23 2024 at 12:05 UTC