bjorn3 opened PR #6043 from less_seq_nums
to main
:
This allows renumbering to be localized to a single block where previously it could affect the entire function. Also saves 32bit of overhead per block.
I originally set out to remove sequence numbers entirely to save 32bit of overhead for every instruction, but the dominator tree code uses sequence numbers to find the relative order of instructions within a block.
Builds on top of https://github.com/bytecodealliance/wasmtime/pull/6042. I split both of them to simplify bisection and help with reviewing.
bjorn3 updated PR #6043 from less_seq_nums
to main
.
cfallin submitted PR review.
bjorn3 updated PR #6043 from less_seq_nums
to main
.
cfallin merged PR #6043.
Last updated: Dec 23 2024 at 13:07 UTC