[llvm-branch-commits] [llvm] [DAG] Explicitly truncate splat operand in get_active_lane_mask expansion (PR #230056)

[via llvm-branch-commits]

================

----------------
Andarwinux wrote:

> I switched it over to mask.beforefirst because I'm not sure if the dependence.war.mask intrinsic will ever be emitted by the LV for x86. Added v2/v3 coverage in [9b2045d](https://github.com/llvm/llvm-project/commit/9b2045dee158f310c0b0da5c4a8686e6b3ee5c30). AVX512 needs the widening TODO to be addressed first.

https://godbolt.org/z/oT8KqMvs1

With -force-partial-aliasing-vectorization -tail-folding-policy=prefer-fold-tail enabled, LV indeed emits it. x86 may consider tail-folded epilogue after #208764, so coverage for this is definitely meaningful.

https://github.com/llvm/llvm-project/pull/230056