[all-commits] [llvm/llvm-project] 5dfba5: [DAG] Fold shuffle(bop(shuffle(x, y), shuffle(z, w)), ...

[Simon Pilgrim via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 5dfba562dd247f731528448ee83785b099f93629
      https://github.com/llvm/llvm-project/commit/5dfba562dd247f731528448ee83785b099f93629
  Author: Simon Pilgrim <llvm-dev at redking.me.uk>
  Date:   2021-02-16 (Tue, 16 Feb 2021)

  Changed paths:
    M llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
    M llvm/lib/Target/X86/X86ISelLowering.cpp
    M llvm/test/CodeGen/X86/bitcast-and-setcc-256.ll
    M llvm/test/CodeGen/X86/promote-cmp.ll

  Log Message:
  -----------
  [DAG] Fold shuffle(bop(shuffle(x,y),shuffle(z,w)),bop(shuffle(a,b),shuffle(c,d)))

Fold shuffle(bop(shuffle(x,y),shuffle(z,w)),bop(shuffle(a,b),shuffle(c,d))) -> bop(shuffle(x,y),shuffle(z,w)),bop(shuffle(a,b),shuffle(c,d))

Attempt to fold from a shuffle of a pair of binops to a binop of shuffles, as long as one/both of the binop sources are also shuffles that can be merged with the outer shuffle. This should guarantee that we remove one binop without introducing any additional shuffles.

Technically there's potential for a merged shuffle's lowering to be poorer than the original shuffle, but it could also be better, and I'm not seeing any regressions as long as we keep the 'don't merge splats' rule already present in MergeInnerShuffle.

This expands and generalizes an existing X86 combine and attempts to merge either of each binop's sources (with an on-the-fly commutation of the shuffle mask) - we couldn't do that in the x86 version as it had to stay in a form that DAGCombine's MergeInnerShuffle would still recognise.

Differential Revision: https://reviews.llvm.org/D96345