| Issue |
154238
|
| Summary |
Missed optimization: fold away neutral fadd 0.0 before fabs
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
zxt5
|
```
define double @src(double %arg0, float %arg1) {
%v2 = fadd float %arg1, 0.000000e+00
%v3 = call noundef float @llvm.fabs.f32(float %v2)
%v4 = fpext float %v3 to double
%v5 = fcmp olt double %arg0, %v4
%v6 = select i1 %v5, double %v4, double %arg0
ret double %v6
}
define double @tgt(double %arg0, float %arg1) {
%v2 = call noundef float @llvm.fabs.f32(float %arg1)
%v3 = fpext float %v2 to double
%v4 = fcmp olt double %arg0, %v3
%v5 = select i1 %v4, double %v3, double %arg0
ret double %v5
}
```
Alive2: https://alive2.llvm.org/ce/z/WyK7Q9
Godbolt: https://godbolt.org/z/n36zdahPM
Pattern found in https://github.com/dtcxzyw/llvm-opt-benchmark/blob/main/bench/opencv/optimized/ts_func.ll
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
