Delete a bunch of uses of getType in favor of TensorOptions. by ezyang · Pull Request #11087 · pytorch/pytorch

ezyang · 2018-08-30T15:46:28Z

Delete a bunch of uses of getType in favor of TensorOptions.

Differential Revision: D9581560

Stacked on #11080

Differential Revision: D9581560 Differential Version: 56526196

Differential Revision: D9581560 Differential Version: 56532263

Differential Revision: D9581560 Differential Version: 56532377

Differential Revision: D9581560 Differential Version: 56535272

Differential Revision: D9581560 Differential Version: 56537640

Differential Revision: D9581560 Differential Version: 56538492

goldsborough

Some suggestions

torch/csrc/cuda/Module.cpp

  using namespace torch::autograd;
  HANDLE_TH_ERRORS
-  Variable var = VariableType::getType(CPU(kByte))->tensor();
+  Variable var = torch::empty({0}, at::device(at::kCPU).dtype(at::kByte));


torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp

    for (auto i = 0; i < numTensors_; i++) {
-      deviceGuard.set_index(i % numDevices_);
-      inputs_[i] = type.tensor({16, 16});
+      inputs_[i] = at::empty({16, 16}, at::device({at::kCUDA, i % numDevices_}).dtype(at::kFloat));


torch/lib/c10d/test/ProcessGroupGlooTest.cpp

    // Initialize tensor list
    std::vector<at::Tensor> tensors = {
-        at::ones({16, 16}, at::TensorOptions(at::CPU(at::kFloat))),
+        at::ones({16, 16}, at::device(at::kCPU).dtype(at::kFloat)),


torch/lib/c10d/test/ProcessGroupGlooTest.cpp

  for (auto i = 0; i < size; i++) {
    auto tensor =
-        at::ones({16, 16}, at::TensorOptions(at::getType(b, at::kFloat))) * i;
+        at::ones({16, 16}, at::TensorOptions(b).dtype(at::kFloat)) * i;


torch/lib/c10d/test/ProcessGroupGlooTest.cpp

          }
          inputs[k][l] =
-              at::ones({16, 16}, at::TensorOptions(type)) * (k * stride + l);
+              at::ones({16, 16}, at::TensorOptions(b).dtype(at::kFloat)) * (k * stride + l);


torch/lib/c10d/test/ProcessGroupNCCLTest.cpp

    for (auto i = 0; i < numDevices_; ++i) {
      deviceGuard.set_index(i);
-      inputs_[i] = type.tensor({3, 3});
+      inputs_[i] = at::empty({3, 3}, at::device(at::kCUDA).dtype(at::kFloat));


torch/lib/c10d/test/ProcessGroupNCCLTest.cpp

      outputs_[i].resize(worldSize_ * numDevices_);
      for (auto j = 0; j < worldSize_ * numDevices_; ++j) {
-        outputs_[i][j] = type.tensor({3, 3});
+        outputs_[i][j] = at::empty({3, 3}, at::device(at::kCUDA).dtype(at::kFloat));


torch/csrc/Generator.cpp

  HANDLE_TH_ERRORS
  THGenerator *generator = THPGenerator_TH_CData(self);
-  Variable var = VariableType::getType(CPU(kByte))->tensor();
+  Variable var = torch::empty({0}, at::device(at::kCPU).dtype(at::kByte));


test/cpp/api/serialization.cpp


      auto x = torch::ones(
-          {5, 5}, torch::getType(torch::Backend::CPU, static_cast<torch::Dtype>(i)));
+          {5, 5}, at::device(at::kCPU).dtype(static_cast<torch::Dtype>(i)));


aten/src/ATen/test/scalar_test.cpp

  if(at::hasCUDA()) {
-    auto & CUDAFloat = C.getType(Backend::CUDA,ScalarType::Float);
-    auto t2 = zeros({4,4}, CUDAFloat);
+    auto t2 = zeros({4,4}, at::device(at::kCUDA).dtype(at::kFloat));


Differential Revision: D9581560 Differential Version: 56544238

Differential Revision: D9581560 Differential Version: 56544398

aten/src/ATen/test/scalar_test.cpp

  if(at::hasCUDA()) {
-    auto & CUDAFloat = C.getType(Backend::CUDA,ScalarType::Float);
-    auto t2 = zeros({4,4}, CUDAFloat);
+    auto t2 = zeros({4,4}, at::kCUDA);


torch/lib/c10d/test/ProcessGroupGlooTest.cpp

        at::DeviceGuard deviceGuard;
        for (auto l = 0; l < stride; l++) {
-          if (type.is_cuda()) {
+          if (b == at::Backend::CUDA) { // NB:wouldn't work with sparse


Summary: Pull Request resolved: pytorch/pytorch#11087 Reviewed By: cpuhrsch Differential Revision: D9581560 fbshipit-source-id: ebe3c4c0956da8a7215ada287bf6526dbcb2b07d

Summary: Pull Request resolved: pytorch#11087 Reviewed By: cpuhrsch Differential Revision: D9581560 fbshipit-source-id: ebe3c4c0956da8a7215ada287bf6526dbcb2b07d

V2: Initial commit

0fec60e

Differential Revision: D9581560 Differential Version: 56526196

ezyang requested review from apaszke, colesbury, ebetica, gchanan, goldsborough, pietern, soumith, teng-li and zdevito as code owners August 30, 2018 15:46

ezyang added 3 commits August 30, 2018 09:45

V3: Use more idiomatic TensorOptions.

097607f

Differential Revision: D9581560 Differential Version: 56532263

V4: Fix error in scalar_test

848c307

Differential Revision: D9581560 Differential Version: 56532377

V5: Fix Backend conversion problem

478016a

Differential Revision: D9581560 Differential Version: 56535272

ezyang mentioned this pull request Aug 30, 2018

Rename getType to getVariableTypeFromBaseType / getVariableType #11095

Closed

ezyang added 2 commits August 30, 2018 10:35

V6: CPU to kCPU

ea28b81

Differential Revision: D9581560 Differential Version: 56537640

V7: Fix more GlooTest problems

ecb8635

Differential Revision: D9581560 Differential Version: 56538492

goldsborough reviewed Aug 30, 2018

View reviewed changes

ezyang added 2 commits August 30, 2018 11:32

V8: CR

5a70f00

Differential Revision: D9581560 Differential Version: 56544238

V9: one more I missed

9c27b23

Differential Revision: D9581560 Differential Version: 56544398

goldsborough approved these changes Aug 30, 2018

View reviewed changes

gchanan reviewed Aug 31, 2018

View reviewed changes

facebook-github-bot closed this in c836a04 Aug 31, 2018

ezyang added the merged label Jun 26, 2019

Conversation

ezyang commented Aug 30, 2018

Uh oh!

goldsborough left a comment

Choose a reason for hiding this comment

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

This comment was marked as off-topic.

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants