[PATCH] [X86][AVX2] vpslldq/vpsrldq byte shifts for AVX2
Simon Pilgrim
llvm-dev at redking.me.uk
Fri Feb 13 02:55:20 PST 2015
The AVX1 splits fail because the x86 computeZeroableShuffleElements() call isn't recursive - so can't peek through nodes (mainly shuffles but concat / subvector etc, as well do crop up). I'm hoping to give that code an overhaul in the near future - plus there is nothing x86 specific about it and would be useful to provide for others targets, the DAGCombiner etc. to check for zeros/zeroables. If you want I can do that work first and then update this patch?
REPOSITORY
rL LLVM
http://reviews.llvm.org/D7596
EMAIL PREFERENCES
http://reviews.llvm.org/settings/panel/emailpreferences/
More information about the llvm-commits
mailing list