[Mlir-commits] [flang] [mlir] [mlir][CSE] Introduce hoist-pure-ops logic to CSE pass (PR #180556)

[lonely eagle]

================

----------------
linuxlonelyeagle wrote:

There is a example:
```
func.func @cse_multiple_regions(%c: i1, %t: tensor<5xf32>) -> (tensor<5xf32>, tensor<5xf32>) {
  %r1 = scf.if %c -> (tensor<5xf32>) {
    %0 = tensor.empty() : tensor<5xf32>
    scf.yield %0 : tensor<5xf32>
  } else {
    scf.yield %t : tensor<5xf32>
  }
  %r2 = scf.if %c -> (tensor<5xf32>) {
    %0 = tensor.empty() : tensor<5xf32>
    scf.yield %0 : tensor<5xf32>
  } else {
    scf.yield %t : tensor<5xf32>
  }
  return %r1, %r2 : tensor<5xf32>, tensor<5xf32>
}
```
if we don't host `op`, if will be 
```
  %0 = tensor.empty() : tensor<5xf32>
  %r1 = scf.if %c -> (tensor<5xf32>) {
    scf.yield %0 : tensor<5xf32>
  } else {
    scf.yield %t : tensor<5xf32>
  }
  %r2 = scf.if %c -> (tensor<5xf32>) {
    scf.yield %0 : tensor<5xf32>
  } else {
    scf.yield %t : tensor<5xf32>
  }
  return %r1, %r2 : tensor<5xf32>, tensor<5xf32>
}
```
That's because it happens before deleting all the dead operations.It will be
```
 %0 = tensor.empty() : tensor<5xf32>  //`exiting`
  %r1 = scf.if %c -> (tensor<5xf32>) {
    scf.yield %0 : tensor<5xf32>
  } else {
    scf.yield %t : tensor<5xf32>
  }
  %r2 = scf.if %c -> (tensor<5xf32>) {
   %1 = tensor.empty() : tensor<5xf32> // `op`
    scf.yield %0 : tensor<5xf32>
  } else {
    scf.yield %t : tensor<5xf32>
  }
  return %r1, %r2 : tensor<5xf32>, tensor<5xf32>
}
```

https://github.com/llvm/llvm-project/pull/180556