[Mlir-commits] [mlir] [mlir][dataflow] Add visitBranchRegionArgument interface to SparseBackwardDataFlowAnalysis and apply it in LivenessAnalysis/RemoveDeadValues (PR #169816)

[Matthias Springer]

================
@@ -166,25 +165,6 @@ void LivenessAnalysis::visitBranchOperand(OpOperand &operand) {
           blocks.push_back(&block);
       }
     }
-
-    // In the block of the successor block argument of RegionBranchOpInterface,
----------------
matthias-springer wrote:

This comment explicitly mentions the IV of `scf.for`. Shouldn't this also cover the IV of `affine.for`? Why is this broken for `affine.for` but not `scf.for`?

I tried running this IR (without your patch):
```mlir
// RUN: mlir-opt %s -test-liveness-analysis

func.func @affine_for() {
  %c1 = arith.constant 1 : index
  %alloc = memref.alloc() : memref<10xindex>
  affine.for %arg0 = 0 to 79 {
    memref.store %c1, %alloc[%c1] : memref<10xindex>
  } {tag = "affine.for"}
  return
}

func.func @scf_for(%lb: index, %ub: index, %step: index) {
  %c1 = arith.constant 1 : index
  %alloc = memref.alloc() : memref<10xindex>
  scf.for %arg0 = %lb to %ub step %step {
    memref.store %c1, %alloc[%c1] : memref<10xindex>
  } {tag = "scf.for"}
  return
}
```

And this is the output that I am seeing:
```
test_tag: affine.for:
 region: #0:
   argument: #0: not live    <-- affine.for IV
test_tag: scf.for:
 operand #0: live
 operand #1: live
 operand #2: live
 region: #0:
   argument: #0: live        <-- scf.for IV
```

Can you make sense of this?


https://github.com/llvm/llvm-project/pull/169816