[PATCH] D41233: [InstCombine] Canonizing 'and' before 'shl'

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a comment.

At least the bswap part of this (and probably handling the regression that @craig.topper pointed out) should be split into independent patches. I don't know what the InstSimplify part is doing yet, but that seems like it should be an independent patch too.

You can see the bswap limitation with a small test like this:

  declare void @extra_use(i32)
    
  define i32 @test7_and_first(i32 %x) {
    %shl = shl i32 %x, 16
    %shr = lshr i32 %x, 16
    %or = or i32 %shl, %shr
    %t1 = and i32 %or, 16711935
    %shl3 = shl nuw i32 %t1, 8
    %and4 = lshr i32 %or, 8
    %shr5 = and i32 %and4, 16711935
    %or6 = or i32 %shl3, %shr5
    ret i32 %or6
  }
  
  define i32 @test7_and_first_extra_use(i32 %x) {
    %shl = shl i32 %x, 16
    %shr = lshr i32 %x, 16
    %or = or i32 %shl, %shr
    %t1 = and i32 %or, 16711935
    %shl3 = shl nuw i32 %t1, 8
    %and4 = lshr i32 %or, 8
    %shr5 = and i32 %and4, 16711935
    %or6 = or i32 %shl3, %shr5
    call void @extra_use(i32 %t1)
    ret i32 %or6
  }
  
  
  define i32 @test7_shl_first(i32 %x) {
    %shl = shl i32 %x, 16
    %shr = lshr i32 %x, 16
    %or = or i32 %shl, %shr
    %and2 = shl i32 %or, 8
    %shl3 = and i32 %and2, -16711936
    %and4 = lshr i32 %or, 8
    %shr5 = and i32 %and4, 16711935
    %or6 = or i32 %shl3, %shr5
    ret i32 %or6
  }



  $ ./opt -instcombine -S bswap.ll 
  
  declare void @extra_use(i32)
  
  define i32 @test7_and_first(i32 %x) {
    %or6 = call i32 @llvm.bswap.i32(i32 %x)
    ret i32 %or6
  }
  
  define i32 @test7_and_first_extra_use(i32 %x) {
    %shl = shl i32 %x, 16
    %shr = lshr i32 %x, 16
    %or = or i32 %shl, %shr
    %t1 = and i32 %or, 16711935
    %shl3 = shl nuw i32 %t1, 8
    %and4 = lshr i32 %or, 8
    %shr5 = and i32 %and4, 16711935
    %or6 = or i32 %shl3, %shr5
    call void @extra_use(i32 %t1)
    ret i32 %or6
  }
  
  define i32 @test7_shl_first(i32 %x) {
    %or6 = call i32 @llvm.bswap.i32(i32 %x)
    ret i32 %or6
  }


Repository:
  rL LLVM

https://reviews.llvm.org/D41233