| Issue |
95216
|
| Summary |
Converting float16 -> float32 in computations seem to behave inconsistently
|
| Labels |
|
| Assignees |
|
| Reporter |
DenisYaroshevskiy
|
Hi!
In the following example tst1 and tst2 produce different assembly, specifically tst2 does `__truncsfhf2` on every add.
```
_Float16 tst1(_Float16 a, _Float16 b, _Float16 c) {
return a + b + c;
}
_Float16 tst2(_Float16 a, _Float16 b, _Float16 c) {
_Float16 res = 0;
res *= a;
res *= b;
res *= c;
return res;
}
```
https://godbolt.org/z/WKh8v1EaE
Is this an expected/desired behaviour? If so - what is the rule governing _Float16 emulation.
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
