| Issue |
60969
|
| Summary |
[MLIR] Invalid IRs generated by irrelevant optimization pass
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
sweead
|
This is a very interesting test case. I found that there are some passes on it that output invalid IRs. There is no `scf` dialect in this test case, so `-scf-for-to-while` should not take effect, but actually produces invalid IRs with no content in funcOp.
I also found several similar bugs on this test case. Maybe the reason behind this problem may be similar, I will try to find the root cause of the bugs.
Reproduce cmd:
```
mlir-opt -scf-for-to-while core-ops.mlir
```
IR:
```
func.func @affine_apply1() {
%i = "arith.constant"() {value = 0: index} : () -> index
%j = "arith.constant"() {value = 1: index} : () -> index
%a = "affine.apply" (%i) { map = affine_map<(d0) -> (d0 + 1)> } :
(index) -> (index)
%b = affine.apply affine_map<()[x] -> (x+1)>()[%i]
return
}
```
Outputs:
```
module {
func.func @affine_apply1() {
return
}
}
```
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs