On f32/f64.rem

[dcodeIO]

I know these are not real instructions, but while implementing a compiler that is not C/C++-based it turned out to be a nuisance. Duplicating fmod every time doesn't seem right to me when embedders could easily provide their anyway ubiquitous implementation.

Could it be re-considered for binary size / peace of mind / alignment with other web-things?

[binji]

WebAssembly typically doesn't provide numeric instructions for operations that aren't commonly available on hardware. That's why there isn't a f32.sin or f32.cos either, though they would probably be useful. Perhaps musl's implementation would be of use: https://github.com/bpowers/musl/blob/master/src/math/fmod.c

[dcodeIO]

That's about what I am doing currently, yeah. I totally understand the exclusion of sin and cos because these can be implemented in various ways yielding more or less precise results depending on requirements.

The absense of f32/f64.rem however creates a situation where a supposedly exact operation, that is widely supported among other web-platform things (here: JS) as an operator, has to be bundled over and over again, even though the embedder could easily provide it. It's the (perceived unnecessary, despite technical arguments) redundancy this introduces that doesn't seem right to me.

[binji]

Well, if you're interested in championing the feature, I'd suggest presenting a proposal with some data at the next CG meeting. That's coming up pretty soon, but even something preliminary may be worthwhile to get some feedback on how to move forward.

[vmukhachev]

can be implemented using fma, for some values:

fma(trunc(x / y), -y, x)