[GitHub] [incubator-tvm] icemelon9 commented on a change in pull request #4775: conv3d_ndhwc schedule

[GitBox <gi...@apache.org>]

icemelon9 commented on a change in pull request #4775: conv3d_ndhwc schedule
URL: https://github.com/apache/incubator-tvm/pull/4775#discussion_r373643228
 
 

 ##########
 File path: topi/python/topi/x86/conv3d.py
 ##########
 @@ -0,0 +1,92 @@
+# 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=invalid-name, unused-variable, too-many-locals
+# pylint: disable=unused-argument, redefined-builtin, no-else-return
+"""Conv3D operators"""
+import tvm
+from .. import generic, tag
+
+@generic.schedule_conv3d_ndhwc.register("cpu")
+def schedule_conv3d_ndhwc(outs):
+    """TOPI schedule callback for conv3d
+
+    Parameters
+    ----------
+    outs: Array of Tensor
+        The computation graph description of conv3d
+        in the format of an array of tensors.
+
+    Returns
+    -------
+    s: Schedule
+        The computation schedule for conv3d.
+    """
+    s = tvm.create_schedule([x.op for x in outs])
+    output_op = outs[0].op
+    scheduled_ops = []
+
+    def traverse(op):
+        """Traverse operators from computation graph"""
+        # inline all one-to-one-mapping operators except the last stage (output)
+        if tag.is_broadcast(op.tag):
+            if op not in s.outputs:
+                s[op].compute_inline()
+            else: # inject custom schedule
+                if len(op.axis) == 5:
+                    # schedule bias + bn + activation
+                    n, d, h, w, c = op.axis
+                    fused = s[op].fuse(n, d, h, w)
+                    s[op].parallel(fused)
+                    s[op].vectorize(c)
+            for tensor in op.input_tensors:
+                if isinstance(tensor.op, tvm.tensor.ComputeOp) and tensor.op not in scheduled_ops:
+                    traverse(tensor.op)
+
+        if 'conv3d_ndhwc' in op.tag:
+            conv = op.output(0)
+            kernel = op.input_tensors[1]
+            # dilation stage
+            if isinstance(kernel.op, tvm.tensor.ComputeOp) and "dilate" in kernel.op.tag:
+                s[kernel].compute_inline()
+
+            # padding stage
+            data = op.input_tensors[0]
+            data_pad = None
+            if isinstance(data.op, tvm.tensor.ComputeOp) and "pad" in data.op.tag:
+                # fuse pad h and w
+                data_pad = data
+                data = data_pad.op.input_tensors[0]
+                _, _, h_pad, w_pad, _ = data_pad.op.axis
+                pad_fused = s[data_pad].fuse(h_pad, w_pad)
+                s[data_pad].parallel(pad_fused)
+
+            # compute conv
+            C = conv
+            n, d, h, w, c = s[C].op.axis
+            s[C].vectorize(c)
+            if op != output_op: # fuse bias + bn + activation
+                _, _, _, _, c_out = output_op.axis
+                s[C].compute_at(s[output_op], c_out)
+            else:
+                # fuse batch, depth, height axes
+                fused = s[C].fuse(n, d, h)
+                s[C].parallel(fused)
+
+        scheduled_ops.append(op)
+
+    traverse(output_op)
 
 Review comment:
   You can use traverse_inline function to avoid redundant inlining broadcast ops.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services