[PATCH] D63992: [InstCombine] Y - ~X --> X + Y + 1 fold (PR42457)

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a reviewer: efriedma.
spatel added a comment.

Not sure which way to go here - a 'not' is considered a less complex op than arbitrary xor/add, and 'not' might be better for value tracking and subsequent instcombines. But the adds are reassociable/commutable.

This might warrant a DAGCombiner patch to avoid regressions before we make the IR decision.

x86 scalar looks better with add+inc:

  movl	%edi, %eax
  notl	%esi
  subl	%esi, %eax

vs.

  leal	1(%rsi,%rdi), %eax

But AArch64 and PowerPC64LE appear to benefit from the 'not' with vector code:

  mvn	v1.16b, v1.16b
  sub	v0.4s, v0.4s, v1.4s

vs.

  add	v0.4s, v1.4s, v0.4s
  movi	v1.4s, #1
  add	v0.4s, v0.4s, v1.4s

And PPC:

  xxlnor 35, 35, 35
  vsubuwm 2, 2, 3

vs.

  vspltisw 4, 1
  vadduwm 2, 3, 2
  vadduwm 2, 2, 4

If we do go in this direction, do we want to increment 1st to reduce the burden on -reassociation? Assuming it will do:
(x + y) + 1 --> (x + 1) + y
...we might as well produce that directly?

cc'ing Eli in case he sees any other motivations to consider.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63992/new/

https://reviews.llvm.org/D63992