alexcrichton opened PR #11084 from alexcrichton:x64-shrink to bytecodealliance:main:
This PR has two commits which culminate in migrating
leato the new assembler. The main trickiness aboutleais that we lower alliaddnodes toleabut rely on an optimization to turnleaback intoaddwhere possible. The first commit here adds infrastructure to use the results of register allocation to refine instruction selection and then theleacommit builds on top of that and uses it. There are a few other examples of using the results of register allocation such asAND RAX, imm32which ISLE never otherwise generates but can be selected after register allocation for a slightly more compact encoding.
alexcrichton requested cfallin for a review on PR #11084.
alexcrichton requested wasmtime-compiler-reviewers for a review on PR #11084.
alexcrichton requested dicej for a review on PR #11084.
alexcrichton requested wasmtime-core-reviewers for a review on PR #11084.
alexcrichton requested abrown for a review on PR #11084.
alexcrichton updated PR #11084.
abrown submitted PR review:
Nice! It looks like there are small code-size wins everywhere and there are probably opportunities for more of this kind of micro-optimization as well.
alexcrichton updated PR #11084.
alexcrichton has enabled auto merge for PR #11084.
alexcrichton merged PR #11084.
Last updated: Dec 06 2025 at 06:05 UTC