!2037 Bug: Fix pytorch infershape origin shape wrong

Merge pull request !2037 from zhaoxinxin/master

ge/graph/passes/infershape_pass.cc

@@ -228,19 +228,13 @@ bool InferShapePass::SameTensorDesc(const GeTensorDescPtr &src, const GeTensorDe
 }
 
 graphStatus InferShapePass::UpdateTensorDesc(const GeTensorDescPtr &src, GeTensorDescPtr &dst, bool &changed) {
-  changed = !SameTensorDesc(src, dst);
-  // refresh src itself
-  src->SetOriginShape(src->GetShape());
-  src->SetOriginDataType(src->GetDataType());
-  TensorUtils::SetRealDimCnt(*src, static_cast<uint32_t>(src->GetOriginShape().GetDims().size()));
-  vector<pair<int64_t, int64_t>> src_shape_range;
-  src->GetShapeRange(src_shape_range);
-  src->SetOriginShapeRange(src_shape_range);
-
-  if (!changed) {
+  changed = false;
+  if (SameTensorDesc(src, dst)) {
     GELOGD("Peer dst tensor_desc is same as src tensor_desc. No need update.");
     return SUCCESS;
   }
+
+  changed = true;
   UpdateShapeAndDType(src, dst);
   GELOGD(
       "UpdatePeerInputDesc from src Node: shape: [%s], datatype: %s, original datatype is %s."