torch.aten.view.dtype from f8E4M3FNUZ to si8 with dynamic dimension fails to lower

[sogartar]

What happened?

    %0 = torch.aten.view.dtype %arg0, %int1
        : !torch.vtensor<[?,32,8,128],f8E4M3FNUZ>, !torch.int
        -> !torch.vtensor<[?,32,8,128],si8>

results in error

small-reproducer.mlir:5:10: error: 'flow.tensor.bitcast' op value set has 1 dynamic dimensions but only 0 dimension values are attached
    %0 = torch.aten.view.dtype %arg0, %int1
         ^
small-reproducer.mlir:5:10: note: see current operation: %1 = "flow.tensor.bitcast"(%0) <{operandSegmentSizes = array<i32: 1, 0, 0>}> : (tensor<?x32x8x128xf8E4M3FNUZ>) -> tensor<?x32x8x128xi8>

Steps to reproduce your issue

Compile the below IR with

iree-compile small-reproducer.mlir \
  -o=small-reproducer.vmfb \
  --iree-hip-target=gfx942 \
  --iree-hal-target-device=hip

func.func @main(
  %arg0: !torch.vtensor<[?,32,8,128],f8E4M3FNUZ>
) -> !torch.vtensor<[?,32,8,128],si8> {
    %int1 = torch.constant.int 1
    %0 = torch.aten.view.dtype %arg0, %int1
        : !torch.vtensor<[?,32,8,128],f8E4M3FNUZ>, !torch.int
        -> !torch.vtensor<[?,32,8,128],si8>
  return %0 : !torch.vtensor<[?,32,8,128],si8>
}

What component(s) does this issue relate to?

MLIR, Compiler

Version information

8fab35c

Additional context

This was encountered when trying to compile Lllam f8 and the bitcasting was required to appease Torch for not supporting torch.Tensor.index_copy_ for torch.float8_e4m3fnuz.
More context at nod-ai/shark-ai#896 (comment)