[PATCH] D82105: [ARM][MVE] Tail-Prediction: revert multiple get.active.lane.mask intrinsics

Sjoerd Meijer via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Thu Jun 18 09:13:28 PDT 2020


SjoerdMeijer created this revision.
SjoerdMeijer added reviewers: samparker, dmgreen, efriedma.
Herald added subscribers: danielkiss, hiraditya, kristof.beyls.
Herald added a project: LLVM.

If the loop is rejected for tail-predication, intrinsic @get.active.lane.mask needs to be reverted to an icmp. We were expecting only 1 occurrence of get.active.lane.mask in the vector body, because that's what the vectoriser generates, but when interleaving is requested multiple @get.active.lane.mask instructions can occur, as shown with the added test case `interleave4`. This change rejects loops with multiple @get.active.lane.masks in them, this can be addressed as a follow up, but importantly for now revert all @get.active.lane.mask to icmps in order to avoid error:

  Cannot select: intrinsic %llvm.get.active.lane.mask"


https://reviews.llvm.org/D82105

Files:
  llvm/lib/Target/ARM/MVETailPredication.cpp
  llvm/test/CodeGen/Thumb2/LowOverheadLoops/basic-tail-pred.ll
  llvm/test/CodeGen/Thumb2/LowOverheadLoops/tail-pred-const.ll
  llvm/test/CodeGen/Thumb2/LowOverheadLoops/tail-reduce.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D82105.271735.patch
Type: text/x-patch
Size: 28904 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200618/2d61568b/attachment.bin>


More information about the llvm-commits mailing list