Golang 1.17.3 performance regression

[cohix]

Hello,

This is not an issue with Wasmtime, I am just hoping to get confirmation of something I'm seeing on my own machine.

It seems there has been a regression in Go 1.17.3 that slows down execution with Wasmtime:

Go 1.17.2:

RUST: 1056
TINYGO: 1026
ASSEMBLYSCRIPT: 1052
GRAIN: 1140
SWIFT: 1775

Go 1.17.3:

RUST: 1737
TINYGO: 1636
ASSEMBLYSCRIPT: 1683
GRAIN: 1888
SWIFT: 2876

(numbers are time in MS to execute 100k runs of a basic helloworld module)

As you can see, there is a noticable time difference after upgrading to 1.17.3.

Would anyone on the wasmtime team be willing to attempt some benchmarks on these two Go versions (I'm sure yours are more sophisticated) to help determine if there is a real performance regression in Go 1.17.3 ?

[alexcrichton]

Oh dear this sounds bad, thanks for the report! We don't have any benchmarks for this embedding right now, so can you gist what code you used to generate the numbers you're getting here?

A quick scan of the 1.17.3 changelog doesn't reveal any culprits unfortunately.

[srenatus]

1.18 is out. Have you noticed anything different on that?