[llvm] [LV][AArch64] Prefer Fixed over Scalable if cost-model is equal (Neoverse V2) (PR #95819)

Paul Walker via llvm-commits llvm-commits at lists.llvm.org
Tue Jun 18 02:47:59 PDT 2024


paulwalker-arm wrote:

Considering the rational as explained in the second paragraph where I see phrases like "predicates does not come entirely for free", "NEON tends to generate slightly less code" and "SVE codegen is slightly less optimal" I'm apposed to having this flags.  If there's other rational then great but to me this looks like a hack to work round cost model deficiencies and generally poor code generation, both of which should be fixed rather than hidden. 

https://github.com/llvm/llvm-project/pull/95819


More information about the llvm-commits mailing list