| Issue |
179494
|
| Summary |
AMDGPU fails to form BFE with 16-bit patterns
|
| Labels |
backend:AMDGPU,
missed-optimization
|
| Assignees |
|
| Reporter |
arsenm
|
These functions should be able to equivalently use v_bfe_u32, but the 16-bit version does not:
https://godbolt.org/z/n5EG4Mh3z
```
; v_bfe_u32 v0, v0, 4, 4
define i32 @bfe_32(i32 %a) {
%shr.i = lshr i32 %a, 4
%and = and i32 %shr.i, 15
ret i32 %and
}
; v_lshrrev_b16_e32 v0, 4, v0
; v_and_b32_e32 v0, 15, v0
define i16 @bfe_16(i16 %a) {
%shr.i = lshr i16 %a, 4
%and = and i16 %shr.i, 15
ret i16 %and
}
```
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
