x64: Lower bitcast, fabs, and fneg in ISLE

[elliottt]

Migrate the bitcast, fabs, and fneg instructions to ISLE

[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.

[cfallin]

r+ on new commits here; will approve once we resolve #4722 and rebase this

[cfallin]

I wonder, would it be better to use a VCodeConstant here to get the appropriate MSBs-only-in-each-lane value?

[elliottt]

Would that be a savings over the two instructions?

[cfallin]

Possibly, though looking at xorps I see the basic SSE version is not reg, reg/mem but just reg, reg so it would still be a separate load; one load vs. dependent chain of two 1-cycle ops, the latter likely wins. Anyway it's an unimportant enough question that I don't care too much to investigate further :-)

[cfallin]

LGTM, thanks!