What is the feature or code improvement you would like to do in Cranelift?

In bytecodealliance/cranelift#1347 I added a temporary lowering for Wasm's load_splat to two Cranelift instructions, load + splat. This generates extra instructions that could be removed by a specialized Cranelift load_splat instruction or by smarter codegen (e.g. complex addressing on splat).

What is the value of adding this in Cranelift?

Fewer instructions produced.

Do you have an implementation plan, and/or ideas for data structures or algorithms to use?

Seeking feedback on which way to proceed: specialized load_splat or smarter codegen.

Have you considered alternative implementations? If so, how are they better or worse than your proposal?

See above.