[llvm] [AMDGPU] Prefer vector i8s in PHI Nodes (PR #91016)

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Wed May 8 08:10:55 PDT 2024


arsenm wrote:

Intrinsically it makes no sense for us to prefer i8 anything. It's a 32-bit architecture. Everything is a 32-bit register. Vectors of i32 should be the canonically preferred type for anything wide. Why would we want i8? 

https://github.com/llvm/llvm-project/pull/91016


More information about the llvm-commits mailing list