[llvm] [AArch64] Set MaxInterleaving to 4 for Neoverse V2 (PR #100385)
Sjoerd Meijer via llvm-commits
llvm-commits at lists.llvm.org
Tue Oct 8 01:28:45 PDT 2024
sjoerdmeijer wrote:
This now includes lowering the minimum epilogue vectorisation factor from 16 to 8 for the Neoverse V2, which mostly recovers the one regression we spotted in cam4_r. A bit of TTI plumbing and new EpilogueVectorizationMinVF hook is introduced to make this now target specific.
This patch now depends on #108190: it needs the `multiplier` information that is now passed into the profitability calculation:
bool LoopVectorizationCostModel::isEpilogueVectorizationProfitable(
const ElementCount VF) const {
const ElementCount VF, const unsigned Multiplier)
https://github.com/llvm/llvm-project/pull/100385
More information about the llvm-commits
mailing list