Remove template parameter from Tensor (#9939)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/9939

Pull Request resolved: https://github.com/facebookresearch/weakly-supervised-action-detection/pull/13

Pull Request resolved: https://github.com/pytorch/translate/pull/166

Pull Request resolved: https://github.com/pytorch/pytorch/pull/9125

Closes https://github.com/pytorch/pytorch/pull/9125

Use inheritance for polymorphism, and remove template parameter
This is to change the templating in call sites, the core implementations will change later

Before Caffe2 Tensor class was compile-time fixed to bind to a particular device/context. With this change, we're making it a runtime property (stored inside the tensor), but preserve the same semantics. For example, one has to specify device type in order to create a Tensor - there are no uninitialized tensors. More specifically the changes are:

1. We added an extra argument *DeviceType* to most of the constructors of the tensor, e.g. (Tensor(DeviceType type)),
2. Semantics of constructor Tensor(const Tensor<SrcContext>& src, ContextForCopy* context); is changed, in this constructor, the second context is passed in to enable us to call the templated Copy function, it could be in a different context as source and target previously, now we'll enforce that the context should have same device type as src, if it is provided.
3. To preserve 'get-or-construct' semantics of Blob, we added specialized getter Blob::GetMutableTensor that verifies both that Blob contains a Tensor and that it's of a correct type
4. Specifically, Tensor type is not default-constructible any more (as we don't have unknown device tensors) and thus some of the code handling STL containers needs to change

Note: Some changes are postponed just to keep this diff a bit smaller. Please see `TODO`s.

Reviewed By: ezyang, houseroad

Differential Revision: D9024330

fbshipit-source-id: e0b8295d2dc6ebe2963383ded5af799ad17164ba

modules/detectron/group_spatial_softmax_op.h

@@ -68,7 +68,7 @@ class GroupSpatialSoftmaxGradientOp final : public Operator<Context> {
  protected:
   int num_classes_;
   StorageOrder order_;
-  Tensor<Context> sum_probs_;
+  Tensor sum_probs_{Context::GetDeviceType()};
 };
 
 } // namespace caffe2

modules/detectron/select_smooth_l1_loss_op.h

@@ -45,7 +45,7 @@ class SelectSmoothL1LossOp final : public Operator<Context> {
   float beta_; // Transition point from L1 to L2 loss
   float scale_; // Scale the loss by scale_
   int dim_; // dimension for 1 anchor prediction
-  Tensor<Context> buff_; // Buffer for element-wise differences
+  Tensor buff_{Context::GetDeviceType()}; // Buffer for element-wise differences
 };
 
 template <typename T, class Context>
@@ -69,7 +69,7 @@ class SelectSmoothL1LossGradientOp final : public Operator<Context> {
   float beta_; // Transition point from L1 to L2 loss
   float scale_; // Scale the loss by scale_
   int dim_; // dimension for 1 anchor prediction
-  Tensor<Context> buff_; // Buffer for element-wise differences
+  Tensor buff_{Context::GetDeviceType()}; // Buffer for element-wise differences
 };
 
 } // namespace caffe2

modules/detectron/sigmoid_cross_entropy_loss_op.h

@@ -44,9 +44,9 @@ class SigmoidCrossEntropyLossOp final : public Operator<Context> {
  protected:
   float scale_;
   int normalize_;
-  Tensor<Context> losses_;
-  Tensor<Context> counts_;
-  Tensor<Context> normalizer_;
+  Tensor losses_{Context::GetDeviceType()};
+  Tensor counts_{Context::GetDeviceType()};
+  Tensor normalizer_{Context::GetDeviceType()};
 };
 
 template <typename T, class Context>
@@ -69,8 +69,8 @@ class SigmoidCrossEntropyLossGradientOp final : public Operator<Context> {
  protected:
   float scale_;
   int normalize_;
-  Tensor<Context> counts_;
-  Tensor<Context> normalizer_;
+  Tensor counts_{Context::GetDeviceType()};
+  Tensor normalizer_{Context::GetDeviceType()};
 };
 
 } // namespace caffe2

modules/detectron/sigmoid_focal_loss_op.h

@@ -47,8 +47,8 @@ class SigmoidFocalLossOp final : public Operator<Context> {
   int num_classes_;
   float gamma_;
   float alpha_;
-  Tensor<Context> losses_;
-  Tensor<Context> counts_;
+  Tensor losses_{Context::GetDeviceType()};
+  Tensor counts_{Context::GetDeviceType()};
 };
 
 template <typename T, class Context>
@@ -74,8 +74,8 @@ class SigmoidFocalLossGradientOp final : public Operator<Context> {
   int num_classes_;
   float gamma_;
   float alpha_;
-  Tensor<Context> counts_;
-  Tensor<Context> weights_; // unignored weights
+  Tensor counts_{Context::GetDeviceType()};
+  Tensor weights_{Context::GetDeviceType()}; // unignored weights
 };
 
 } // namespace caffe2

modules/detectron/smooth_l1_loss_op.h

@@ -44,7 +44,7 @@ class SmoothL1LossOp final : public Operator<Context> {
  protected:
   float beta_; // Transition point from L1 to L2 loss
   float scale_; // Scale the loss by scale_
-  Tensor<Context> buff_; // Buffer for element-wise differences
+  Tensor buff_{Context::GetDeviceType()}; // Buffer for element-wise differences
 };
 
 template <typename T, class Context>
@@ -67,7 +67,7 @@ class SmoothL1LossGradientOp final : public Operator<Context> {
  protected:
   float beta_; // Transition point from L1 to L2 loss
   float scale_; // Scale the loss by scale_
-  Tensor<Context> buff_; // Buffer for element-wise differences
+  Tensor buff_{Context::GetDeviceType()}; // Buffer for element-wise differences
 };
 
 } // namespace caffe2

modules/detectron/softmax_focal_loss_op.h

@@ -52,7 +52,7 @@ class SoftmaxFocalLossOp final : public Operator<Context> {
   float alpha_;
   int num_classes_;
   StorageOrder order_;
-  Tensor<Context> losses_;
+  Tensor losses_{Context::GetDeviceType()};
 };
 
 template <typename T, class Context>
@@ -83,7 +83,7 @@ class SoftmaxFocalLossGradientOp final : public Operator<Context> {
   float alpha_;
   int num_classes_;
   StorageOrder order_;
-  Tensor<Context> buff_;
+  Tensor buff_{Context::GetDeviceType()};
 };
 
 } // namespace caffe2