Make formats less SIMD-specific

[abrown]

This renames and slightly alters several formats:

• InsertLane becomes TernaryImm8
• ExtractLane becomes BinaryImm8
• BinaryImm becomes BinaryImm64 for consistency

This has additional comments in #1762 about the ordering of operands; in changing InsertLane to TernaryImm8 I changed the order of the operands in the CLIF IR (value, imm, value to value, value, imm) so that this format will make more sense for other instructions (x86_pshufd, x86_pblendw). Other than that this should be a straightforward renaming.

A heads up at the Cranelift API level: users of Cranelift IR (e.g. @bjorn3, @jyn514?) might notice this naming chaning if they are using Format or InstructionData directly.

[abrown]

Waiting on #1762...

[github-actions]

Subscribe to Label Action

cc @bnjbvr

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

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

• bnjbvr: cranelift

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

Learn more.

[jyn514]

I am not using the textual IR for anything besides debugging, so no complaints from me :)

[github-actions]

Subscribe to Label Action

cc @fitzgen

This issue or pull request has been labeled: "cranelift:area:peepmatic"

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

• fitzgen: cranelift:area:peepmatic

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

Learn more.

[bjorn3]

I am currently using neither insertlane nor extractlane. I also don't use Format or InstructionData either.

[iximeow]

This look good! and learning why it is the way it is has been informative:

(InsertLane) Why change op1, op1-mod, op2-style IR to op1, op2, op1-mod-style?

Because some instructions, like x86's pshufw, would want a three-arg-one-immediate format, but it would be src, imm, dest, where there's no relation between imm and any operand other than the shuffling pshufw actually performs. "philosophically speaking" in how Cranelift IR is defined, existing formats just aren't quite a good fit.

(InsertLane) Why not have both op1, op1-mod, op2 and op1, op2, mod formats?

The instruction format machinery is lax about where immediates show up. This means that Cranelift can't disceren between op, imm, op and op, op, imm formats

Consequently, having formats that are less SIMD-oriented makes sense both for insert/extract and other instructions. And maybe one unlikely day we'll reuse the ternary format for 3-arg imul :)

edit: unwelcome CI failure though, with no logs?

[abrown]

unwelcome CI failure though, with no logs?

Yeah, I have no idea. I re-ran them and everything was fine 🤷.