[SDAG] Miscompilation at -O3

[dtcxzyw]

Reproducer:

; bin/lli reduced.ll
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

define i32 @main() {
entry:
  br label %func_188.exit.i.i

func_188.exit.i.i:                                ; preds = %func_188.exit.i.i, %entry
  %or659.i167180.i.i = phi i32 [ 0, %entry ], [ 1, %func_188.exit.i.i ]
  %conv48.i.i = trunc i32 %or659.i167180.i.i to i8
  %and.i.i.i = and i32 %or659.i167180.i.i, 1
  %tobool80.not.i.i.i = icmp eq i32 %and.i.i.i, 0
  br i1 %tobool80.not.i.i.i, label %if.else.i.i.i, label %func_188.exit.i.i

if.else.i.i.i:                                    ; preds = %func_188.exit.i.i
  %cmp183.i.i.i = icmp sgt i8 %conv48.i.i, 0
  %ext = zext i1 %cmp183.i.i.i to i32 
  ret i32 %ext
}

The expected return value is 0. But I got 1.
llvm version: 1bb8db5

[llvmbot]

@llvm/issue-subscribers-backend-x86

Author: Yingwei Zheng (dtcxzyw)

Reproducer: ``` ; bin/lli reduced.ll target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128" target triple = "x86_64-unknown-linux-gnu"

define i32 @main() {
entry:
br label %func_188.exit.i.i

func_188.exit.i.i: ; preds = %func_188.exit.i.i, %entry
%or659.i167180.i.i = phi i32 [ 0, %entry ], [ 1, %func_188.exit.i.i ]
%conv48.i.i = trunc i32 %or659.i167180.i.i to i8
%and.i.i.i = and i32 %or659.i167180.i.i, 1
%tobool80.not.i.i.i = icmp eq i32 %and.i.i.i, 0
br i1 %tobool80.not.i.i.i, label %if.else.i.i.i, label %func_188.exit.i.i

if.else.i.i.i: ; preds = %func_188.exit.i.i
%cmp183.i.i.i = icmp sgt i8 %conv48.i.i, 0
%ext = zext i1 %cmp183.i.i.i to i32
ret i32 %ext
}

The expected return value is 0. But I got 1.
llvm version: 1bb8db5d5e1a3ea99525fd2648164a257955cd6d
</details>

[dtcxzyw]

@RKSimon Unfortunately I cannot find the bug. Now I am going to bed. Can you please have a look?

[RKSimon]

trunk codegen:

main:                                   # @main
        xorl    %eax, %eax
.LBB0_1:                                # %func_188.exit.i.i
        testb   $1, %al
        movl    $1, %eax
        jne     .LBB0_1
        setge   %al
        movzbl  %al, %eax
        retq

20.x codegen:

main:                                   # @main
        xorl    %eax, %eax
.LBB0_1:                                # %func_188.exit.i.i
        movl    %eax, %ecx
        movl    $1, %eax
        testb   $1, %cl
        jne     .LBB0_1
        xorl    %eax, %eax
        testb   %cl, %cl
        setg    %al
        retq

[RKSimon]

CC @phoebewang as I think this is related to #155586 / #156002

[phoebewang]

CC @phoebewang as I think this is related to #155586 / #156002

Yes, testb is back with #156002