[Relay][Op] Pad operator (#1843)

docs/langref/relay_op.rst

@@ -55,6 +55,7 @@ This level enables typical convnet models.
    tvm.relay.nn.global_avg_pool2d
    tvm.relay.nn.upsampling
    tvm.relay.nn.batch_flatten
+   tvm.relay.nn.pad
    tvm.relay.nn.lrn
    tvm.relay.nn.l2_normalize
 

include/tvm/relay/attrs/nn.h

@@ -223,8 +223,19 @@ struct UpSamplingAttrs : public tvm::AttrsNode<UpSamplingAttrs> {
   }
 };
 
+/*! \brief Attributes used for the padding operator */
+struct PadAttrs : public tvm::AttrsNode<PadAttrs> {
+  double pad_value;
+  Array<Array<IndexExpr> > pad_width;
 
-
+  TVM_DECLARE_ATTRS(PadAttrs, "relay.attrs.PadAttrs") {
+    TVM_ATTR_FIELD(pad_value).set_default(0.0)
+      .describe("Specifies the strides of the convolution.");
+    TVM_ATTR_FIELD(pad_width)
+      .describe("Number of values padded to the edges of each axis, "
+                "in the format of ((before_1, after_1), ..., (before_N, after_N))");
+  }
+};
 
 /*! \brief Attributes for LRN operator */
 struct LRNAttrs : public tvm::AttrsNode<LRNAttrs> {

python/tvm/relay/op/nn/nn.py

@@ -429,7 +429,6 @@ def batch_flatten(data):
     """
     return _make.batch_flatten(data)
 
-
 def relu(data):
     """Rectified linear unit.
 
@@ -449,6 +448,32 @@ def relu(data):
     return _make.relu(data)
 
 
+def pad(data,
+        pad_width,
+        pad_value=0.0):
+    r"""Padding
+
+    This operator takes in a tensor and pads each axis by the specified
+    widths using the specified value.
+
+    Parameters
+    ----------
+    data: relay.Expr
+        The input data to the operator
+    pad_width: tuple of <tuple of <int>>, required
+        Number of values padded to the edges of each axis, in the format
+        of ((before_1, after_1), ..., (before_N, after_N))
+    pad_value: float, optional, default=0.0
+        The value used for padding
+
+    Returns
+    -------
+    result : relay.Expr
+        The computed result.
+    """
+    return _make.pad(data, pad_width, pad_value)
+
+
 def lrn(data, size=5, axis=1, bias=2, alpha=.00001, beta=0.75):
     """This operator takes data as input and does local response normalization.
 
@@ -484,9 +509,9 @@ def lrn(data, size=5, axis=1, bias=2, alpha=.00001, beta=0.75):
     result : relay.Expr
         The computed result.
     """
-
     return _make.lrn(data, size, axis, alpha, beta, bias)
 
+
 def l2_normalize(data, eps, axis=None):
     """Perform L2 normalization on the input data
 

src/relay/op/nn/pad.cc

@@ -0,0 +1,86 @@
+/*!
+ *  Copyright (c) 2018 by Contributors
+ * \file pad.cc
+ * \brief Implementation of operator pad
+ */
+#include <tvm/ir_operator.h>
+#include <tvm/relay/op.h>
+#include <tvm/relay/attrs/nn.h>
+#include <vector>
+#include "layout.h"
+
+namespace tvm {
+namespace relay {
+
+TVM_REGISTER_NODE_TYPE(PadAttrs);
+
+bool PadRel(const Array<Type>& types,
+            int num_inputs,
+            const Attrs& attrs,
+            const TypeReporter& reporter) {
+  CHECK_EQ(types.size(), 2);
+  const auto* data = types[0].as<TensorTypeNode>();
+  if (data == nullptr) return false;
+
+  const PadAttrs* param = attrs.as<PadAttrs>();
+  CHECK(param != nullptr);
+
+  // check that pad widths match lengths
+  CHECK(data->shape.size() == param->pad_width.size())
+    << "There should be as many pad width pairs as shape dimensions "
+    << "but the shape has " << data->shape.size() << " dimensions "
+    << "and there are " << param->pad_width.size() << " pad width pairs.";
+
+  // each pad width element should be a pair of positive integers
+  std::vector<IndexExpr> oshape;
+  for (size_t i = 0; i < param->pad_width.size(); i++) {
+    CHECK(param->pad_width[i].size() == 2)
+      << "Each pad width element should be a pair but at index " << i
+      << " there are " << param->pad_width[i].size() << " elements.";
+
+    auto width1 = as_const_int(param->pad_width[i][0]);
+    auto width2 = as_const_int(param->pad_width[i][1]);
+    CHECK(width1 != nullptr);
+    CHECK(width2 != nullptr);
+
+    CHECK(*width1 >= 0)
+      << "Param width elements should be positive but first pad width at "
+      << "index " << i << " is " << *width1 << ".";
+    CHECK(*width2 >= 0)
+      << "Param width elements should be positive but first pad width at "
+      << "index " << i << " is " << *width2 << ".";
+
+    auto padding = make_const(data->shape[i].type(), *width1 + *width2);
+    oshape.push_back(data->shape[i] + padding);
+  }
+
+  reporter->Assign(types[1], TensorTypeNode::make(Array<IndexExpr>(oshape),
+                                                  data->dtype));
+  return true;
+}
+
+// Handler to create a call to the padding op used by front-end FFI
+  Expr MakePad(Expr data, Array<Array<IndexExpr> > pad_width, double pad_value) {
+  auto attrs = make_node<PadAttrs>();
+  attrs->pad_value = pad_value;
+  attrs->pad_width = std::move(pad_width);
+  static const Op& op = Op::Get("nn.pad");
+  return CallNode::make(op, {data}, Attrs(attrs), {});
+}
+
+TVM_REGISTER_API("relay.op.nn._make.pad")
+.set_body([](const TVMArgs& args, TVMRetValue* rv) {
+    runtime::detail::unpack_call<Expr, 3>(MakePad, args, rv);
+  });
+
+RELAY_REGISTER_OP("nn.pad")
+.describe(R"code(Pad for n-D tensor.
+
+)code" TVM_ADD_FILELINE)
+.set_num_inputs(1)
+.add_argument("data", "Tensor", "The input tensor.")
+.set_support_level(2)
+.add_type_rel("Pad", PadRel);
+
+}  // namespace relay
+}  // namespace tvm

tests/python/relay/test_op_level2.py

@@ -196,10 +196,35 @@ def test_flatten_infer_type():
     ftype = func.checked_type
     assert ftype.ret_type == relay.ty.TensorType((d1, ((2*d3)*3)), "float32")
 
+def test_pad_infer_type():
+    # entirely concrete case
+    ib = relay.ir_builder.IRBuilder()
+    n, c, h, w = 1, 2, 3, 4
+    t = ib.param("t", relay.TensorType((n, c, h, w), "float32"))
+    with ib.function(t) as func:
+        ib.ret(relay.nn.pad(t.var, ((1, 1), (2, 2), (3, 3), (4, 4))))
+    ib.ret(func)
+    func = relay.ir_pass.infer_type(ib.env, func.to_func())
+    ftype = func.checked_type
+    assert ftype.ret_type == relay.TensorType((3, 6, 9, 12), "float32")
+
+    # some symbolic values
+    ib = relay.ir_builder.IRBuilder()
+    n, c, h, w = tvm.var("n"), 2, 3, tvm.var("w")
+    t = ib.param("t", relay.TensorType((n, c, h, w), "float32"))
+    with ib.function(t) as func:
+        ib.ret(relay.nn.pad(t.var, ((1, 1), (2, 2), (3, 3), (4, 4))))
+    ib.ret(func)
+    func = relay.ir_pass.infer_type(ib.env, func.to_func())
+    ftype = func.checked_type
+    assert ftype.ret_type == relay.TensorType((n + 2, 6, 9, w + 8), "float32")
+
 
 if __name__ == "__main__":
     test_conv2d_infer_type()
     test_pool2d_infer_type()
     test_upsampling_infer_type()
     test_flatten_infer_type()
+    test_pad_infer_type()
     test_conv2d_transpose_infer_type()
+