Skip to content

Conversation

@hrydgard
Copy link
Owner

@hrydgard hrydgard commented Jun 9, 2024

Don't break up lwl/lwr/swl/swr instructions into components when using the interpreter, it can handle the original instructions faster directly.

Remove an unnecessary mask operation when using ins some ways.

Optimize away redundant store+load vector operations (seen a lot in the Wipeout games).

None of these alone have a big effect, but if we keep shaving the size of the generated IR code down, performance will improve. Plus, these are fun to work on and some of this will also benefit the IR Jits.

@hrydgard hrydgard added the IRInterpreter Occurs with IR Interpreter but not with another CPU backend. label Jun 9, 2024
@hrydgard hrydgard added this to the v1.18.0 milestone Jun 9, 2024
@hrydgard hrydgard merged commit 5dec3ca into master Jun 9, 2024
@hrydgard hrydgard deleted the ir-lwr branch June 9, 2024 07:48
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

IRInterpreter Occurs with IR Interpreter but not with another CPU backend.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants