Back out "[pytorch][PR] Support upsample" (pytorch#13413)

Summary:
Pull Request resolved: pytorch#13413

Original commit changeset: d5db200365f1

Reviewed By: houseroad

Differential Revision: D12870356

fbshipit-source-id: be115d2370636786901c822895664ccace2a9bc2

Author: zrphercule2

test/onnx/expect/TestOperators.test_upsample.expect

@@ -2,29 +2,23 @@ ir_version: 3
 producer_name: "pytorch"
 producer_version: "0.4"
 graph {
-  node {
-    output: "1"
-    op_type: "Constant"
-    attribute {
-      name: "value"
-      t {
-        dims: 4
-        data_type: FLOAT
-        raw_data: "\000\000\200?\000\000\200?\000\000\000@\000\000\000@"
-      }
-      type: TENSOR
-    }
-  }
   node {
     input: "0"
-    input: "1"
-    output: "2"
+    output: "1"
     op_type: "Upsample"
     attribute {
       name: "mode"
       s: "linear"
       type: STRING
     }
+    attribute {
+      name: "scales"
+      floats: 1
+      floats: 1
+      floats: 2
+      floats: 2
+      type: FLOATS
+    }
   }
   name: "torch-jit-export"
   input {
@@ -50,7 +44,7 @@ graph {
     }
   }
   output {
-    name: "2"
+    name: "1"
     type {
       tensor_type {
         elem_type: FLOAT

test/onnx/test_operators.py

@@ -75,6 +75,7 @@ def assertONNX(self, f, args, params=None, **kwargs):
             import test_onnx_common
             model_def = onnx.ModelProto.FromString(onnx_model_pb)
             onnx.checker.check_model(model_def)
+
             if _onnx_test:
                 test_function = inspect.stack()[1][0].f_code.co_name
                 test_name = test_function[0:4] + "_operator" + test_function[4:]

torch/onnx/symbolic.py

@@ -642,10 +642,8 @@ def replication_pad(g, input, padding):
 def upsample_nearest2d(g, input, output_size):
     height_scale = float(output_size[-2]) / input.type().sizes()[-2]
     width_scale = float(output_size[-1]) / input.type().sizes()[-1]
-    scales = g.op("Constant", value_t=torch.tensor([1., 1., height_scale,
-                                                    width_scale]))
-
-    return g.op("Upsample", input, scales,
+    return g.op("Upsample", input,
+                scales_f=[1., 1., height_scale, width_scale],
                 mode_s="nearest")
 
 
@@ -655,9 +653,8 @@ def upsample_bilinear2d(g, input, output_size, align_corners):
         return _unimplemented("upsample_bilinear2d", "align_corners == True")
     height_scale = float(output_size[-2]) / input.type().sizes()[-2]
     width_scale = float(output_size[-1]) / input.type().sizes()[-1]
-    scales = g.op("Constant", value_t=torch.tensor([1., 1., height_scale,
-                                                    width_scale]))
-    return g.op("Upsample", input, scales,
+    return g.op("Upsample", input,
+                scales_f=[1., 1., height_scale, width_scale],
                 mode_s="linear")