[llvm] [NVPTX] Support copysign PTX instruction (PR #107800)

Artem Belevich via llvm-commits llvm-commits at lists.llvm.org
Mon Sep 9 16:35:17 PDT 2024


Artem-B wrote:

>  it looks like ptxas is doing a better job then the SelectionDAG (2 vs. 3 instructions).

`2x LOP3.LUT` look like a likely win over `SHF, ISETP, FSEL`.

We should probably start using `lop3` more often, as it could potentially handle other common operations a bit better than the discrete logical ops we use now.




https://github.com/llvm/llvm-project/pull/107800


More information about the llvm-commits mailing list