Optimize more reduction-of-an-extend cases

[scottmcm]

I was exploring ireduce patterns more after I removed them from #7693 (comment) at fitzgen's request.

In doing so, I noticed that there were some simpler cases missing first, like ireduce.i8 sextend.i32 my_i16 → ireduce.i8 my_i16 and ireduce.i16 sextend.i32 my_i8 → sextend.i16 my_i8.

So this PR adds those (and the unsigned equivalents).

[github-actions]

Subscribe to Label Action

cc @cfallin, @fitzgen

Details

This issue or pull request has been labeled: "cranelift", "isle"

Thus the following users have been cc'd because of the following labels:

• cfallin: isle
• fitzgen: isle

To subscribe or unsubscribe from this label, edit the .github/subscribe-to-label.json configuration file.

Learn more.

[fitzgen]

Thanks for splitting this out!

Would you mind enabling the trace-log feature and looking at this trace log when running wasmtime compile sightglass/benchmarks/spidermonkey/benchmark.wasm¹ and comparing the numbers before and after this PR? I want to make sure we aren't accidentally blowing up the size of the egraph.

Footnotes

1. github.com/bytecodealliance/sightglass ↩

[scottmcm]

I don't really know what I'm doing, but here's what I got: 7711.diff.txt

I ran .\target\release\wasmtime.exe compile ..\spidermonkey_benchmark.wasm -D log-to-files -C parallel-compilation=n before and after, which meant all the logs were in "wasmtime.dbg.main".

Looks plausible? Nothing that seems like a huge blow-up, but a bunch of places with a couple more CLIF instructions that end up as fewer lowered vcode instructions

[fitzgen]

Stats look good -- thanks for getting those!

[alexcrichton]

The first failure above was spurious, but the second failure looks like it might be related to this PR

[scottmcm]

Yup, conflicted with myself from #7719. Should be good now.