[Mlir-commits] [mlir] [mlir] Make emitWarning/emitRemark to use OpWithFlags (PR #217804)

[lonely eagle]

linuxlonelyeagle wrote:

> Can you show before/after for a typical case?

Here is a clear example from my setup. Each remark represents the result of an analysis on the loop structure. As you can see, emitting remarks directly on the loop creates far too much output.

Before
```
a.mlir:4:3: remark: message0
  affine.for %i = 0 to %arg1 {
  ^
a.mlir:4:3: note: see current operation: 
affine.for %arg2 = 0 to %arg1 {
  affine.for %arg3 = 1 to %arg1 {
    %0 = affine.load %arg0[%arg3, %arg2] : memref<?x?xf32>
    %1 = affine.load %arg0[%arg2, %arg3 - 1] : memref<?x?xf32>
    %2 = arith.addf %0, %0 : f32
    affine.store %2, %arg0[%arg2, %arg3] : memref<?x?xf32>
  }
}
a.mlir:4:3: remark: message1
  affine.for %i = 0 to %arg1 {
  ^
a.mlir:4:3: note: see current operation: 
affine.for %arg2 = 0 to %arg1 {
  affine.for %arg3 = 1 to %arg1 {
    %0 = affine.load %arg0[%arg3, %arg2] : memref<?x?xf32>
    %1 = affine.load %arg0[%arg2, %arg3 - 1] : memref<?x?xf32>
    %2 = arith.addf %0, %0 : f32
    affine.store %2, %arg0[%arg2, %arg3] : memref<?x?xf32>
  }
}
a.mlir:4:3: remark: message2
  affine.for %i = 0 to %arg1 {
  ^
a.mlir:4:3: note: see current operation: 
affine.for %arg2 = 0 to %arg1 {
  affine.for %arg3 = 1 to %arg1 {
    %0 = affine.load %arg0[%arg3, %arg2] : memref<?x?xf32>
    %1 = affine.load %arg0[%arg2, %arg3 - 1] : memref<?x?xf32>
    %2 = arith.addf %0, %0 : f32
    affine.store %2, %arg0[%arg2, %arg3] : memref<?x?xf32>
  }
}
a.mlir:4:3: remark: message3
  affine.for %i = 0 to %arg1 {
  ^
a.mlir:4:3: note: see current operation: 
affine.for %arg2 = 0 to %arg1 {
  affine.for %arg3 = 1 to %arg1 {
    %0 = affine.load %arg0[%arg3, %arg2] : memref<?x?xf32>
    %1 = affine.load %arg0[%arg2, %arg3 - 1] : memref<?x?xf32>
    %2 = arith.addf %0, %0 : f32
    affine.store %2, %arg0[%arg2, %arg3] : memref<?x?xf32>
  }
}
a.mlir:4:3: remark: message4
```

After
```
a.mlir:4:3: remark: message0
  affine.for %i = 0 to %arg1 {
  ^
a.mlir:4:3: note: see current operation: affine.for %arg2 = 0 to %arg1 {...}
a.mlir:4:3: remark: message1
  affine.for %i = 0 to %arg1 {
  ^
a.mlir:4:3: note: see current operation: affine.for %arg2 = 0 to %arg1 {...}
a.mlir:4:3: remark: message2
  affine.for %i = 0 to %arg1 {
  ^
a.mlir:4:3: note: see current operation: affine.for %arg2 = 0 to %arg1 {...}
a.mlir:4:3: remark: message3
  affine.for %i = 0 to %arg1 {
  ^
a.mlir:4:3: note: see current operation: affine.for %arg2 = 0 to %arg1 {...}
a.mlir:4:3: remark: message4
  affine.for %i = 0 to %arg1 {
```

https://github.com/llvm/llvm-project/pull/217804