x64: Improve codegen for vectors with constant shift amounts

[alexcrichton]

I stumbled across this working on #5795 and figured this was a nice opportunity to improve the codegen here.

[alexcrichton]

I'll admit I don't know what's happening here. This I believe is the constant pool area and capstone is trying to decode it as instructions, but I don't know why after this change, which should generate a smaller constant pool, is generating more instructions here. I would hazard a guess that capstone automatically stoped beforehand for some reason and now it's keeping-on-going, but I'm not sure.

[jameysharp]

I think this is at least partially the same issue that @elliottt is tackling in #5798. We've discussed ways to entirely quit trying to disassemble constants since we think the machbuffer knows where they are, but one step at a time.

[jameysharp]

Since #5798 is merged, can you try rebasing and see if the filetests look any better now?

[jameysharp]

Nevermind, I see you already did and it looks like it didn't help. I'm going to declare that this is harmless and move on with reviewing the rest of this PR.

[github-actions]

Subscribe to Label Action

cc @cfallin, @fitzgen

Details

This issue or pull request has been labeled: "cranelift", "cranelift:area:x64", "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.

[jameysharp]

Yup, this looks right to me. Thanks!