| Issue |
115072
|
| Summary |
[EarlyCSE] Optimization ignores specified floating point rounding mode
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
patrick-rivos
|
Testcase:
```llvm ir
define float @test() {
call void @llvm.set.rounding(i32 2) ;; Round up
%a = bitcast i32 1 to float ;; 1e-45
%b = bitcast i32 u0x3F800000 to float ;; 1.0
%res = fadd float %a, %b
ret float %res
}
```
The optimized version is:
```llvm ir
define float @test() {
call void @llvm.set.rounding(i32 2)
ret float 1.000000e+00
}
```
IIUC this ignores the specified rounding mode since it rounds down the result of the fadd.
Godbolt: https://godbolt.org/z/471Ph5EWP
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
