[PATCH] D63963: [Codegen][SelectionDAG] X u% C == 0 fold: non-splat vector improvements

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added a comment.

It really bothered me, so here's alive proofs for powers-of-two:

  $ ~/src/alive2/build-Clang-release/alive /tmp/test.ll 
  OMP: Info #270: omp_set_nested routine deprecated, please use omp_set_max_active_levels instead.
  Processing /tmp/test.ll..
  
  ----------------------------------------
  Name: power-of-two
    %o0 = shl i16 1, %K ; so %o0 is a power-of-two
    %o1 = urem i16 %X, %o0
    %r = icmp eq i16 %o1, 0
  =>
    %o0 = shl i16 1, %K ; so %o0 is a power-of-two
    %o1 = urem i16 %X, %o0
    %n0 = mul i16 %X, 1 ; multiplicative inverse of d0=1 is 1
    %n1 = fshr i16 %n0, %n0, %K
    %n2 = udiv i16 -1, %o0
    %r = icmp ule i16 %n1, %n2
  
  Done: 1
  Optimization is correct!
  
  ----------------------------------------
  Name: power-of-two
    %o0 = shl i16 1, %K ; so %o0 is a power-of-two
    %o1 = urem i16 %X, %o0
    %r = icmp ne i16 %o1, 0
  =>
    %o0 = shl i16 1, %K ; so %o0 is a power-of-two
    %o1 = urem i16 %X, %o0
    %n0 = mul i16 %X, 1 ; multiplicative inverse of d0=1 is 1
    %n1 = fshr i16 %n0, %n0, %K
    %n2 = udiv i16 -1, %o0
    %r = icmp ugt i16 %n1, %n2
  
  Done: 1
  Optimization is correct!
  
  ----------------------------------------
  Name: one
    %o1 = urem i16 %X, 1
    %r = icmp eq i16 %o1, 0
  =>
    %o1 = urem i16 %X, 1
    %n0 = mul i16 %X, 0
    %n1 = fshr i16 %n0, %n0, 0
    %r = icmp ule i16 %n1, -1
  
  Done: 1
  Optimization is correct!
  
  ----------------------------------------
  Name: one
    %o1 = urem i16 %X, 1
    %r = icmp ne i16 %o1, 0
  =>
    %o1 = urem i16 %X, 1
    %n0 = mul i16 %X, 0
    %n1 = fshr i16 %n0, %n0, 0
    %r = icmp ugt i16 %n1, -1
  
  Done: 1
  Optimization is correct!


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63963/new/

https://reviews.llvm.org/D63963