[Mlir-commits] [mlir] [mlir][memref] Add a new `ReifyResultShapes` pass (PR #145927)

[llvmlistbot at llvm.org]

MaheshRavishankar wrote:

Looking at 
```
#map = affine_map<(d0) -> (-d0 + 256)>
func.func @func(%arg0: f32, %arg1: index, %arg2: tensor<64x?x64xf32>) -> tensor<1x?x64xf32> {
  %0 = affine.apply #map(%arg1)
  %extracted_slice = tensor.extract_slice %arg2[0, 0, 0] [1, %arg1, 64] [1, 1, 1] : tensor<64x?x64xf32> to tensor<1x?x64xf32>
  %padded = tensor.pad %extracted_slice low[0, 0, 0] high[0, %0, 0] {
  ^bb0(%arg3: index, %arg4: index, %arg5: index):
    tensor.yield %arg0 : f32
  } : tensor<1x?x64xf32> to tensor<1x?x64xf32>
  return %padded : tensor<1x?x64xf32>
}

// mlir-opt --reify-result-shapes
#map = affine_map<()[s0] -> (-s0 + 256)>
func.func @func(%arg0: f32, %arg1: index, %arg2: tensor<64x?x64xf32>) -> tensor<1x?x64xf32> {
  %0 = affine.apply #map()[%arg1]
  %extracted_slice = tensor.extract_slice %arg2[0, 0, 0] [1, %arg1, 64] [1, 1, 1] : tensor<64x?x64xf32> to tensor<1x?x64xf32>
  %padded = tensor.pad %extracted_slice low[0, 0, 0] high[0, %0, 0] {
  ^bb0(%arg3: index, %arg4: index, %arg5: index):
    tensor.yield %arg0 : f32
  } : tensor<1x?x64xf32> to tensor<1x256x64xf32>
  %cast = tensor.cast %padded : tensor<1x256x64xf32> to tensor<1x?x64xf32>
  return %cast : tensor<1x?x64xf32>
}
```

This has nothing to do with reifying result shapes. This is just a const shape propagation that should be covered by the "cast folding" pattern here https://github.com/llvm/llvm-project/blob/d699fbd2033e291e8d658777d76b3b9ae1652f44/mlir/lib/Dialect/Tensor/IR/TensorOps.cpp#L3394 (or the one below it).

There isnt a tensor.dim in the input IR, so there is nothing to "reify" . Might be a mismatch in what you expect that interface to do. It is explicitly meant to resolve `tensor.dim` operations.

https://github.com/llvm/llvm-project/pull/145927