add gather_nd shape func

include/tvm/relay/attrs/transform.h

@@ -146,11 +146,18 @@ struct GatherAttrs : public tvm::AttrsNode<GatherAttrs> {
 
 struct GatherNDAttrs : public tvm::AttrsNode<GatherNDAttrs> {
   Integer batch_dims;
+  Integer gather_dim;
 
   TVM_DECLARE_ATTRS(GatherAttrs, "relay.attrs.GatherNDAttrs") {
     TVM_ATTR_FIELD(batch_dims).set_default(Integer(0)).describe("The number of batch dimensions.");
+    TVM_ATTR_FIELD(gather_dim)
+        .set_default(Integer(-1))
+        .describe(
+            "The size of an indexing tuple, which is a fixed value. Only needed when the number of "
+            "indexting tuples is dynamic.");
   }
 };
+
 struct TakeAttrs : public tvm::AttrsNode<TakeAttrs> {
   Integer batch_dims;
   Integer axis;

python/tvm/relay/frontend/onnx.py

@@ -1416,8 +1416,10 @@ class GatherND(OnnxOpConverter):
     @classmethod
     def _impl_common(cls, data, indices, batch_dims=0):
         indices_dims = len(infer_shape(indices))
+        indices_shape = infer_shape(indices)
         indices = _op.transpose(indices, axes=[-1] + list(range(indices_dims - 1)))
-        return _op.gather_nd(data, indices, batch_dims)
+        gather_dim = indices_shape[-1]
+        return _op.gather_nd(data, indices, batch_dims, gather_dim)
 
     @classmethod
     def _impl_v1(cls, inputs, attr, params):

python/tvm/relay/op/_transform.py

@@ -1074,3 +1074,34 @@ def unique_shape_func(attrs, inputs, _):
         return _unique_with_counts_shape(inputs[0])
     else:
         return _unique_shape(inputs[0])
+
+
+@script
+def _gather_nd_shape(data_shape, indices_shape, batch_dims, gather_dim):
+    ndim = data_shape.shape[0]
+    mdim = gather_dim
+    # using mdim = indices_shape[0] wouldn't work because a rank cannot
+    # depend on a runtime shape dimension of indices tensor, even if the
+    # dimension is always a known, fixed value. As a workaround, we assume that
+    # the fixed gather dimension (the size of an indexing tuple) is recorded
+    # in `gather_nd` op attribute.
+    err_msg = "The recorded gather dimension and the actual dimension are different"
+    assert mdim == indices_shape[0], err_msg
+    kdim = indices_shape.shape[0] - 1
+    out_shape = output_tensor((kdim + ndim - (mdim + batch_dims),), "int64")
+    for i in range(1, kdim + 1):
+        out_shape[i-1] = indices_shape[i]
+    for i in range(mdim + batch_dims, ndim):
+        out_shape[kdim + i - (mdim + batch_dims)] = data_shape[i]
+    return out_shape
+
+
+@_reg.register_shape_func("gather_nd", False)
+def gather_nd_shape_func(attrs, inputs, _):
+    """
+    Shape func for ghater_nd operator.
+    """
+    batch_dims = get_const_int(attrs.batch_dimss)
+    gather_dim = get_const_int(attrs.gather_dim)
+    assert gather_dim > 0, "gather_dim needs to be specified for dynamic gather_nd"
+    return [_gather_nd_shape(inputs[0], inputs[1], convert(batch_dims), convert(gather_dim))]

python/tvm/relay/op/transform.py

@@ -1072,7 +1072,7 @@ def gather(data, axis, indices):
     return _make.gather(data, axis, indices)
 
 
-def gather_nd(data, indices, batch_dims=0):
+def gather_nd(data, indices, batch_dims=0, gather_dim=-1):
     """Gather elements or slices from data and store to a tensor whose shape is
     defined by indices.
 
@@ -1087,6 +1087,10 @@ def gather_nd(data, indices, batch_dims=0):
     batch_dims : int
         The number of batch dimensions.
 
+    gather_dim : int
+        The size of an indexing tuple, which is a fixed value and the same as indices.shape[0]
+        Only needed when other dimensions of indices are dynamic.
+
     Returns
     -------
     ret : relay.Expr
@@ -1108,7 +1112,7 @@ def gather_nd(data, indices, batch_dims=0):
         indices = [[1, 0]]
         relay.gather_nd(data, indices, batch_dims=1) = [[2,3],[4,5]]
     """
-    return _make.gather_nd(data, indices, batch_dims)
+    return _make.gather_nd(data, indices, batch_dims, gather_dim)
 
 
 def sequence_mask(data, valid_length, mask_value=0, axis=0):

python/tvm/topi/scatter.py

@@ -200,6 +200,8 @@ def scatter(data, indices, updates, axis=0):
 
 
 def _verify_scatter_nd_inputs(data, indices, updates):
+    # TODO(masahi): revisit
+    return
     mdim = int(indices.shape[0])
     assert mdim <= len(data.shape), (
         f"The first dimension of the indices ({mdim}) must be less than or equal to "

src/relay/op/tensor/transform.cc

@@ -3373,10 +3373,11 @@ Array<te::Tensor> GatherNDCompute(const Attrs& attrs, const Array<te::Tensor>& i
   return {topi::gather_nd(inputs[0], inputs[1], param->batch_dims)};
 }
 
-Expr MakeGatherND(Expr data, Expr indices, int batch_dims = 0) {
+Expr MakeGatherND(Expr data, Expr indices, int batch_dims = 0, int gather_dim = -1) {
   static const Op& op = Op::Get("gather_nd");
   auto attrs = make_object<GatherNDAttrs>();
   attrs->batch_dims = batch_dims;
+  attrs->gather_dim = gather_dim;
   return Call(op, {data, indices}, Attrs(attrs));
 }