Jamey Sharp

Now that the x64 backend is migrated to ISLE, is it time to re-visit this optimization? cc: @elliottt

Maybe the test-runner should keep a mapping between heap-allocated names and the "pair of u32" representation, just like an embedder would?

> I think this would be a good idea, if we can remove `TestCase` from `ExternalName` I think the struct size goes down, so it could potentially reduce memory usage...

I was hesitant, but actually yeah, renaming the test functions to fit within the limit is fine. Let's just plan that later we'll do a partial revert to get the...

Well, darn. This seems to have broken `cranelift-fuzzgen`, and I don't understand why. But 20 out of 50 cases I tested from my existing corpus crash now, and it looks...

Okay, I'm reasonably convinced that "Undeclared function u0:0 is referenced by u0:1" has the same root cause as #4757 and #4758, although it's a different symptom. Previously these undefined functions...

I've pushed an alternative on top of this branch, to exactly match the behavior of `Vec::split_off` in `MmapVec`. Would you review it and merge if you're happy with it?

How much of this PR is still applicable today? I see that `XmmLoadConst` still exists in the x64 backend, so if removing that simplifies the backend then it'd be nice...

As I understand it, since #4826 merged, all constants should be aligned now. If there are any instructions that require aligned loads for floats or integers I think those should...

According to #2294, newer versions of Rust shouldn't emit the `__rustc_debug_gdb_scripts_section__` symbol any more. Based on #511, it sounds like 1.50 is new enough, which is over a year old....