torch-blade shape anaysis related problems

[zzpmiracle]

We forcely anaysis sahpe for each op, and erase shape information after failed, to get totally dynamic graph in https://github.com/alibaba/BladeDISC/pull/929 . This issue tracks realted problems occured in examples, TorchBench, and other scenes.

Occured problems:

shape anaysis

• [ ] https://github.com/alibaba/BladeDISC/actions/runs/3840036856/jobs/6538601043#step:6:1651
• [ ] SuperSloMo
  • [ ] aten::index.Tensor(Tensor self, Tensor?[] indices) -> Tensor
  • [ ] aten::upsample_bilinear2d.vec(Tensor input, SymInt[]? output_size, bool align_corners, float[]? scale_factors) -> Tensor
  • [ ] aten::to.dtype_layout(Tensor(a) self, *, ScalarType? dtype=None, Layout? layout=None, Device? device=None, bool? pin_memory=None, bool non_blocking=False, bool copy=False, MemoryFormat? memory_format=None) -> Tensor(a)
  • [ ] aten::stack(Tensor[] tensors, int dim=0) -> Tensor
• [ ] yolov3
  • [ ] aten::meshgrid(Tensor[] tensors) -> Tensor[]
  • [ ] aten::sigmoid_(Tensor(a!) self) -> Tensor(a!)
  • [ ] aten::copy_(Tensor(a!) self, Tensor src, bool non_blocking=False) -> Tensor(a!)
  • [ ] aten::grid_sampler(Tensor input, Tensor grid, int interpolation_mode, int padding_mode, bool align_corners) -> Tensor

related lowering problems

• [x] torch.ones/zeros with dynamic size input not support lowering

[tanyokwok]

ones/zeros can be decomposed into aten.full, I can fix this later