Skip to content

Conversation

lenary
Copy link
Member

@lenary lenary commented Sep 16, 2025

This echoes what has been done for the lmul>1 vector registers, increasing priority for pairs compared to single GPRs.

The changes here are not particularly large, in part due to the fact that GPRPairs are not generated in many examples, but there look to be small savings in the number of register pair copies (where instead a GPR is copied).

This echoes what has been done for the lmul>1 vector registers,
increasing priority for pairs compared to single GPRs.

The changes here are not particularly large, in part due to the fact
that GPRPairs are not generated in many examples, but there look to be
small savings in the number of register pair copies (where instead a GPR
is copied).
Comment on lines -12 to -13
; CHECK-NEXT: mv a0, a2
; CHECK-NEXT: mv a1, a3
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I believe this is the only saving showin in this PR - swapping a GPRPair copy for a GPR copy.

I will investigate further if there are savings on our benchmarks/codebases.

@lenary
Copy link
Member Author

lenary commented Sep 22, 2025

Looks to be achieved similarly with #160084 which also looks less disruptive and more general. Closing

@lenary lenary closed this Sep 22, 2025
@lenary lenary deleted the pr/riscv-gpr-pair-alloc branch September 22, 2025 15:19
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