| Issue |
183337
|
| Summary |
[AMDGPU] Wrong op_sel_hi for v_fma_mix_f32_bf16
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
AlexAUT
|
In this [reproducer](https://godbolt.org/z/zrnEj5j6G) which does `cvt_b16_to_f32(a) + cvt_i8_to_f32(b)` we get:
```
v_fma_mix_f32_bf16 v1, v1, 1.0, v2 op_sel_hi:[1,1,0]
```
Which does compute `v1 = cvt_to_f32(v1) * cvt_to_f32(1.0) + v2` however the inline constant is already `f32` so it should not be converted, therefore I would expect something like:
```
v_fma_mix_f32_bf16 v1, v1, 1.0, v2 op_sel_hi:[1,0,0]
```
where we do not convert the inline constant since it's already f32.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
