Skip to content

Conversation

@zherczeg
Copy link
Collaborator

Automatically target for stores.
LoadMemIdx int/float variants gets different code paths.

@zherczeg zherczeg force-pushed the more_bytecode_update branch 2 times, most recently from 611acda to 7b8ace2 Compare December 18, 2025 10:22
@zherczeg zherczeg force-pushed the more_bytecode_update branch 5 times, most recently from 63341e8 to 628756e Compare January 8, 2026 09:13
@zherczeg zherczeg marked this pull request as draft January 8, 2026 09:14
@zherczeg zherczeg force-pushed the more_bytecode_update branch from 628756e to dbff2b2 Compare January 8, 2026 09:15
Float / int load is split.
Target is computed automatically for stores.
The 32 / 64 bit stores use a different argument order to prevent code merging.

Signed-off-by: Zoltan Herczeg [email protected]
@zherczeg zherczeg force-pushed the more_bytecode_update branch from dbff2b2 to 0bcb8db Compare January 8, 2026 09:23
@zherczeg zherczeg marked this pull request as ready for review January 8, 2026 10:04
@zherczeg
Copy link
Collaborator Author

zherczeg commented Jan 8, 2026

The reason I work on these dst/src swapping is that the "merging code paths" in the interpreter with computed goto is a valid compiler optimization, although no other compiler did it so far. However, it is kind of a minefield, since computed goto address to byte code index conversion is not possible, and better to fix it before other compilers start doing it.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant