Annotate overload information on torchlib functions

[justinchuby]

No description provided.

[BowenBao]

List of unannotated ops found in diagnostics from onnx bench. @justinchuby @titaiwangms
Update: added a few new ones, removed ones that were missing torchlib op instead of overload.

 'aten.add.Tensor': 35510,
 'aten.bitwise_and.Tensor': 12,
 'aten.clamp.Tensor': 2690,
 'aten.div.Tensor': 10622,
 'aten.div.Tensor_mode': 2,
 'aten.empty.memory_format': 12486,
 'aten.eq.Scalar': 72,
 'aten.eq.Tensor': 112,
 'aten.fill.Tensor': 28,
 'aten.ge.Scalar': 4,
 'aten.ge.Tensor': 4,
 'aten.gt.Scalar': 46,
 'aten.le.Tensor': 32,
 'aten.lt.Scalar': 80,
 'aten.masked_fill.Scalar': 360,
 'aten.masked_fill.Tensor': 396,
 'aten.mul.Tensor': 24214,
 'aten.ne.Scalar': 630,
 'aten.pow.Tensor_Scalar': 528,
 'aten.pow.Tensor_Tensor': 1820,
 'aten.rsub.Scalar': 354,
 'aten.scatter_reduce.two': 18,
 'aten.select.int': 4669,
 'aten.slice.Tensor': 17717,
 'aten.split.Tensor': 3182,
 'aten.sub.Tensor': 7868,
 'aten.sum.dim_IntList': 6122,
 'aten.transpose.int': 13219,
 'aten.unbind.int': 1188,
 'aten.where.self': 732

[BowenBao]

@justinchuby oops I'm also working on this

[justinchuby]

aha! do you have a list you are working on? I haven't done anything yet

[BowenBao]

#919

[BowenBao]

There is aten::div.Tensor_Mode left, which should be unsupported, but somehow worked with fallback.

[BowenBao]

There is aten::div.Tensor_Mode left, which should be unsupported, but somehow worked with fallback.

Tracked by pytorch/pytorch#106057, along with other exporter dispatcher nearest match warnings.

[titaiwangms]

There is aten::div.Tensor_Mode left, which should be unsupported, but somehow worked with fallback.

Fallback seems reasonable, given the current implementation. However, there is one perfect match for this op in th sarif you provided in pytorch/pytorch#106057. That seems like a big bug lol.