| Issue |
151303
|
| Summary |
[InstCombine] Incorrect fabs + nsz fold
|
| Labels |
miscompilation,
llvm:instcombine
|
| Assignees |
|
| Reporter |
nikic
|
https://alive2.llvm.org/ce/z/FmBrma
```llvm
define float @src(float %a) {
%i32 = bitcast float %a to i32
%cmp = icmp slt i32 %i32, 0
call void @llvm.assume(i1 %cmp)
%b = fneg nsz float %a
%c = call float @llvm.fabs(float %b)
ret float %c
}
```
The fabs is optimized away here, even though `fneg nsz` can flip the sign bit of zero.
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
