[onnx] handle dynamic padSize tensor in onnx.Pad (#3214)

- Fix pad size to data_rank for dynamic paddingSize Tensor.
- This fix is in accordance with [input
specification](https://onnx.ai/onnx/operators/onnx__Pad.html#inputs) for
onnx.Pad
- Impl will need to be updated for dynamic padSize when support for
`axes` is added.

lib/Conversion/TorchOnnxToTorch/DefaultDomainGtoP.cpp

@@ -1319,9 +1319,18 @@ void mlir::torch::onnx_c::populateDefaultDomainGtoP(
                                              "expect 1-d pad tensor");
 
         int64_t padsSize = padsShape[0];
-        if (padsSize == Torch::kUnknownSize)
-          return rewriter.notifyMatchFailure(binder.op,
-                                             "pad length is unknown");
+        if (padsSize == Torch::kUnknownSize) {
+          // As per onnx.Pad documentation, padSize = 2*num_data_axes
+          // (if axes param not passed). Need to be updated when adding
+          // support for `axes` param.
+          auto dataOpTy = data.getType().cast<Torch::ValueTensorType>();
+          TensorType dataTensor = dataOpTy.toBuiltinTensor();
+          if (!dataTensor || !dataTensor.hasRank())
+            return rewriter.notifyMatchFailure(
+                binder.op, "pad length unknown and data operand unranked");
+          int64_t dataRank = dataTensor.getRank();
+          padsSize = 2 * dataRank;
+        }
 
         Value constantValue;
         if (binder.getNumOperands() >= 3) {