This is a follow up on D139181.
Packing InputChunks going from 88B to 80B.
Packing OutputSegment going from 112B to 104B.
Details
- Reviewers
arichardson
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
lld/wasm/InputChunks.h | ||
---|---|---|
109 | Maybe put this below the last i32 member (outSecOff) and before the first packed value (sectionKind)? Do we have any tooling for automating this kind of stuff? |
lld/wasm/InputChunks.h | ||
---|---|---|
109 | Note that the placement may affect performance as well as code size (mostly for x86 when the assembly may switch from 1-byte to 4-byte displacement). I usually benchmark two runs. I know little about wasm, but I think the field is probably not accessed that frequently. My comment here is a reminder that field reordering should not be done blindly. |
Maybe put this below the last i32 member (outSecOff) and before the first packed value (sectionKind)?
Do we have any tooling for automating this kind of stuff?