[RELAY][DYN] Implementation of the dynamic pad operator (apache#6284)

include/tvm/relay/attrs/nn.h

@@ -85,7 +85,7 @@ struct Conv1DAttrs : public tvm::AttrsNode<Conv1DAttrs> {
         .set_default(NullValue<IndexExpr>());
     TVM_ATTR_FIELD(kernel_size)
         .describe("Specifies the dimensions of the convolution window.")
-        .set_default(NullValue<Array<IndexExpr> >());
+        .set_default(NullValue<Array<IndexExpr>>());
     TVM_ATTR_FIELD(data_layout)
         .set_default("NCW")
         .describe(
@@ -148,7 +148,7 @@ struct Conv2DAttrs : public tvm::AttrsNode<Conv2DAttrs> {
         .set_default(NullValue<IndexExpr>());
     TVM_ATTR_FIELD(kernel_size)
         .describe("Specifies the dimensions of the convolution window.")
-        .set_default(NullValue<Array<IndexExpr> >());
+        .set_default(NullValue<Array<IndexExpr>>());
     TVM_ATTR_FIELD(data_layout)
         .set_default("NCHW")
         .describe(
@@ -242,7 +242,7 @@ struct Conv2DWinogradAttrs : public tvm::AttrsNode<Conv2DWinogradAttrs> {
         .set_default(NullValue<IndexExpr>());
     TVM_ATTR_FIELD(kernel_size)
         .describe("Specifies the dimensions of the convolution window.")
-        .set_default(NullValue<Array<IndexExpr> >());
+        .set_default(NullValue<Array<IndexExpr>>());
     TVM_ATTR_FIELD(data_layout)
         .set_default("NCHW")
         .describe(
@@ -331,7 +331,7 @@ struct Conv3DAttrs : public tvm::AttrsNode<Conv3DAttrs> {
         .set_default(NullValue<IndexExpr>());
     TVM_ATTR_FIELD(kernel_size)
         .describe("Specifies the dimensions of the convolution window.")
-        .set_default(NullValue<Array<IndexExpr> >());
+        .set_default(NullValue<Array<IndexExpr>>());
     TVM_ATTR_FIELD(data_layout)
         .set_default("NCDHW")
         .describe(
@@ -381,7 +381,7 @@ struct Conv3DTransposeAttrs : public tvm::AttrsNode<Conv3DTransposeAttrs> {
             "i.e. the number of output channels in the convolution.");
     TVM_ATTR_FIELD(kernel_size)
         .describe("The dimensions of the convolution window.")
-        .set_default(NullValue<Array<IndexExpr> >());
+        .set_default(NullValue<Array<IndexExpr>>());
     TVM_ATTR_FIELD(strides)
         .set_default(Array<IndexExpr>({1, 1, 1}))
         .describe("The strides of the convolution.");
@@ -480,7 +480,7 @@ struct Conv3DWinogradAttrs : public tvm::AttrsNode<Conv3DWinogradAttrs> {
         .set_default(NullValue<IndexExpr>());
     TVM_ATTR_FIELD(kernel_size)
         .describe("Specifies the dimensions of the convolution window.")
-        .set_default(NullValue<Array<IndexExpr> >());
+        .set_default(NullValue<Array<IndexExpr>>());
     TVM_ATTR_FIELD(data_layout)
         .set_default("NCDHW")
         .describe(
@@ -539,7 +539,7 @@ struct Conv2DTransposeAttrs : public tvm::AttrsNode<Conv2DTransposeAttrs> {
             "i.e. the number of output channels in the convolution.");
     TVM_ATTR_FIELD(kernel_size)
         .describe("The dimensions of the convolution window.")
-        .set_default(NullValue<Array<IndexExpr> >());
+        .set_default(NullValue<Array<IndexExpr>>());
     TVM_ATTR_FIELD(strides)
         .set_default(Array<IndexExpr>({1, 1}))
         .describe("The strides of the convolution.");
@@ -626,7 +626,7 @@ struct Conv1DTransposeAttrs : public tvm::AttrsNode<Conv1DTransposeAttrs> {
             "i.e. the number of output channels in the convolution.");
     TVM_ATTR_FIELD(kernel_size)
         .describe("The dimensions of the convolution window.")
-        .set_default(NullValue<Array<IndexExpr> >());
+        .set_default(NullValue<Array<IndexExpr>>());
     TVM_ATTR_FIELD(strides)
         .set_default(Array<IndexExpr>({1}))
         .describe("The strides of the convolution.");
@@ -1016,7 +1016,7 @@ struct UpSampling3DAttrs : public tvm::AttrsNode<UpSampling3DAttrs> {
 /*! \brief Attributes used for the padding operator */
 struct PadAttrs : public tvm::AttrsNode<PadAttrs> {
   double pad_value;
-  Array<Array<IndexExpr> > pad_width;
+  Array<Array<Integer>> pad_width;
   std::string pad_mode;
 
   TVM_DECLARE_ATTRS(PadAttrs, "relay.attrs.PadAttrs") {
@@ -1037,7 +1037,7 @@ struct PadAttrs : public tvm::AttrsNode<PadAttrs> {
 /*! \brief Attributes used for the MirrorPadding operator */
 struct MirrorPadAttrs : public tvm::AttrsNode<MirrorPadAttrs> {
   std::string mode;
-  Array<Array<IndexExpr> > pad_width;
+  Array<Array<IndexExpr>> pad_width;
 
   TVM_DECLARE_ATTRS(MirrorPadAttrs, "relay.attrs.MirrorPadAttrs") {
     TVM_ATTR_FIELD(mode)
@@ -1242,7 +1242,7 @@ struct DeformableConv2DAttrs : public tvm::AttrsNode<DeformableConv2DAttrs> {
         .set_default(NullValue<IndexExpr>());
     TVM_ATTR_FIELD(kernel_size)
         .describe("Specifies the dimensions of the convolution window.")
-        .set_default(NullValue<Array<IndexExpr> >());
+        .set_default(NullValue<Array<IndexExpr>>());
     TVM_ATTR_FIELD(data_layout)
         .set_default("NCHW")
         .describe(

include/tvm/topi/nn.h

@@ -124,6 +124,8 @@ inline tvm::te::Tensor prelu(const tvm::te::Tensor& x, const tvm::te::Tensor& sl
  * "constant" pads with constant_value;
  * "edge" pads using the edge values of the input array;
  * "reflect" pads by reflecting values with respect to the edges.
+ * \param dyn_output_shape Output shape of the pad op, default nullptr.
+ * You only need to pass this in if the shape was evaluated dynamically.
  * \param name The name of the operation
  * \param tag The tag to mark the operation
  *
@@ -151,30 +153,40 @@ inline tvm::te::Tensor prelu(const tvm::te::Tensor& x, const tvm::te::Tensor& sl
 inline tvm::te::Tensor pad(const tvm::te::Tensor& t, const tvm::Array<tvm::PrimExpr>& pad_before,
                            tvm::Array<tvm::PrimExpr> pad_after = tvm::Array<tvm::PrimExpr>(),
                            PrimExpr pad_value = PrimExpr(), std::string name = "T_pad",
-                           std::string tag = kElementWise, std::string pad_mode = "constant") {
+                           std::string tag = kElementWise, std::string pad_mode = "constant",
+                           const Array<PrimExpr>* dyn_output_shape = nullptr) {
   if (pad_after.size() < pad_before.size()) {
     for (size_t i = pad_after.size(); i < pad_before.size(); ++i) {
       pad_after.push_back(pad_before[i]);
     }
   }
+
   arith::Analyzer analyzer;
   CHECK_GE(pad_before.size(), 1);
   CHECK_EQ(pad_before.size(), pad_after.size());
-  tvm::Array<tvm::PrimExpr> output_shape;
   tvm::Array<tvm::PrimExpr> pad_before_int32;
   tvm::Array<tvm::PrimExpr> pad_after_int32;
+
   for (const auto& ele : pad_before) {
     pad_before_int32.push_back(tvm::cast(tvm::DataType::Int(32), ele));
   }
   for (const auto& ele : pad_after) {
     pad_after_int32.push_back(tvm::cast(tvm::DataType::Int(32), ele));
   }
-  for (size_t i = 0; i < t->shape.size(); ++i) {
-    if (i >= pad_before.size()) {
-      output_shape.push_back(t->shape[i]);
-    } else {
-      output_shape.push_back(
-          analyzer.Simplify(t->shape[i] + pad_before_int32[i] + pad_after_int32[i]));
+
+  tvm::Array<tvm::PrimExpr> output_shape;
+  if (dyn_output_shape == nullptr) {
+    for (size_t i = 0; i < t->shape.size(); ++i) {
+      if (i >= pad_before.size()) {
+        output_shape.push_back(t->shape[i]);
+      } else {
+        output_shape.push_back(
+            analyzer.Simplify(t->shape[i] + pad_before_int32[i] + pad_after_int32[i]));
+      }
+    }
+  } else {
+    for (size_t i = 0; i < dyn_output_shape->size(); i++) {
+      output_shape.push_back((*dyn_output_shape)[i]);
     }
   }
 

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

@@ -0,0 +1,20 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+# pylint: disable=wildcard-import, redefined-builtin, invalid-name
+"""The Relay namespace containing dynamic ops."""
+
+from . import _nn

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

@@ -0,0 +1,20 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+"""Constructor APIs"""
+import tvm._ffi
+
+tvm._ffi._init_api("relay.op.dyn.nn._make", __name__)

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

@@ -0,0 +1,46 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+# pylint: disable=no-else-return, invalid-name, unused-argument, too-many-arguments, consider-using-in
+"""Backend compiler related feature registration"""
+
+from __future__ import absolute_import
+
+from tvm.te.hybrid import script
+from ...op import register_shape_func
+from ...op import register_broadcast_schedule
+
+# pad
+register_broadcast_schedule("dyn.nn.pad")
+
+#####################
+#  Shape functions  #
+#####################
+
+@script
+def _dyn_pad_shape_func(data, pad_width):
+    ndim = len(data.shape)
+    out = output_tensor((ndim,), "int64")
+    for i in const_range(ndim):
+        out[i] = int64(pad_width[i, 0] + pad_width[i, 1] + data.shape[i])
+    return out
+
+@register_shape_func("dyn.nn.pad", True)
+def pad_shape_func(attrs, inputs, data):
+    """
+    Shape function for dynamic pad op.
+    """
+    return [_dyn_pad_shape_func(inputs[0], inputs[1])]

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

@@ -19,7 +19,9 @@
 from tvm.relay import expr
 
 from . import _make
+from ..dyn.nn import _make as _dyn_make
 from .util import get_pad_tuple1d, get_pad_tuple2d, get_pad_tuple3d
+from ...expr import const, Expr
 
 
 def conv1d(data,
@@ -1410,7 +1412,7 @@ def prelu(data, alpha, axis=1):
 
 def pad(data,
         pad_width,
-        pad_value=0.0,
+        pad_value=0,
         pad_mode='constant'):
     r"""Padding
 
@@ -1421,10 +1423,10 @@ def pad(data,
     ----------
     data: tvm.relay.Expr
         The input data to the operator
-    pad_width: tuple of <tuple of <int>>, required
+    pad_width: tuple of <tuple of <int>>, or tvm.relay.Expr, 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
+    pad_value: float, or tvm.relay.Expr, optional, default=0
         The value used for padding
     pad_mode: 'constant', 'edge', 'reflect'
         'constant' pads with constant_value pad_value
@@ -1435,6 +1437,12 @@ def pad(data,
     result : tvm.relay.Expr
         The computed result.
     """
+    if (isinstance(pad_width, Expr) or (isinstance(pad_value, Expr))):
+        if not isinstance(pad_width, Expr):
+            pad_width = const(list(pad_width))
+        if not isinstance(pad_value, Expr):
+            pad_value = const(pad_value)
+        return _dyn_make.pad(data, pad_width, pad_value, pad_mode)
     return _make.pad(data, pad_width, pad_value, pad_mode)
 
 

python/tvm/te/hybrid/calls.py

@@ -73,7 +73,7 @@ def _math_intrin(func_id, args):
     from tvm.tir import op
     return getattr(op, func_id)(*args)
 
-sqrt = log = exp = tanh = sigmoid = power = popcount = _math_intrin #pylint: disable=invalid-name
+sqrt = log = exp = tanh = sigmoid = power = popcount = round = _math_intrin #pylint: disable=invalid-name
 
 
 def _min_max(func_id, args):

python/tvm/te/hybrid/runtime.py

@@ -126,6 +126,7 @@ def max_num_threads(allow_none=True):
     'exp'            : numpy.exp,
     'sigmoid'        : sigmoid,
     'popcount'       : popcount,
+    'round'          : round,
     'likely'         : lambda cond: cond,
     'uint8'          : numpy.uint8,
     'uint16'         : numpy.uint16,

python/tvm/topi/nn/upsampling.py

@@ -57,7 +57,6 @@ def upsampling(data, scale_h, scale_w, layout="NCHW", method='nearest_neighbor',
     elif layout == "NHWC":
         out_shape = (simplify(topi.cast(te.round(data.shape[1] * scale_h), data.shape[1].dtype)),
                      simplify(topi.cast(te.round(data.shape[2] * scale_w), data.shape[2].dtype)))
-
     else:
         raise ValueError("not support this layout {} yet".format(layout))
     coord_trans = "align_corners" if align_corners else "asymmetric"