[llvm-branch-commits] [llvm] [LV] Mask off possibly aliasing vector lanes (PR #100579)

[Sander de Smalen via llvm-branch-commits]

================
@@ -0,0 +1,126 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter-out-after "^middle.block:" --filter-out-after "^scalar.ph:" --version 4
+; RUN: opt -S -mtriple=aarch64-unknown-linux-gnu -mattr=+sve2 -passes=loop-vectorize,instcombine,early-cse -prefer-predicate-over-epilogue=predicate-dont-vectorize -force-vector-interleave=1 %s | FileCheck %s
----------------
sdesmalen-arm wrote:

My recommendation would be not to add too many passes like instcombine/early-cse here, because that may mask issues. It's good to know that CSE will combine the two `llvm.loop.dependence.*.mask` intrinsics (the loop invariant one in the loop body and the one in the check block), but it doesn't necessarily need to check that in the LV test.

https://github.com/llvm/llvm-project/pull/100579