Propagate device data names (#1157)

* Propagate device data names

* Address PR comment

* Add example usage

* Add test for device data names

* Make TorchMlirComputation fields protected

* Add lazy backend device data name unit tests

* Disable lazy backend tests if LTC is disabled

* Add comments

CMakeLists.txt

@@ -52,6 +52,12 @@ endif()
 # TODO: Reenable LTC once OOT build is successful (https://github.com/llvm/torch-mlir/issues/1154)
 option(TORCH_MLIR_ENABLE_LTC "Enables LTC backend" OFF)
 
+if(TORCH_MLIR_ENABLE_LTC)
+    set(ENV{TORCH_MLIR_ENABLE_LTC} 1)
+else()
+    set(ENV{TORCH_MLIR_ENABLE_LTC} 0)
+endif()
+
 torch_mlir_add_llvm_external_project(
   torch-mlir-dialects
   TORCH_MLIR_DIALECTS

python/test/lazy_backend/device_data_name.py

@@ -0,0 +1,44 @@
+# Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+# Also available under a BSD-style license. See LICENSE.
+
+# RUN: %PYTHON %s | FileCheck %s
+
+
+import torch
+import torch._lazy
+
+import torch_mlir.reference_lazy_backend._REFERENCE_LAZY_BACKEND as lazy_backend
+
+from run_test import run_test
+
+lazy_backend._initialize()
+
+device = "lazy"
+
+
+# CHECK: 0 input tensors found
+# -----
+# CHECK: PASS - test_no_device_data_name
+@run_test
+def test_no_device_data_name():
+    x = torch.tensor(1).to(device)
+    y = torch.tensor(2).to(device)
+    z = x + y
+    torch._lazy.mark_step()
+
+
+# CHECK: Input tensor: input_x
+# CHECK: 1 input tensors found
+# -----
+# CHECK: PASS - test_device_data_name
+@run_test
+def test_device_data_name():
+    x = torch.tensor(1).to(device)
+    y = torch.tensor(2).to(device)
+
+    lazy_backend.set_parameter_name(x, "input_x")
+
+    z = x + y
+    torch._lazy.mark_step()

python/test/lazy_backend/run_test.py

@@ -0,0 +1,23 @@
+# Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+# Also available under a BSD-style license. See LICENSE.
+
+# RUN: true
+
+
+def run_test(*args, XPASS=False, XFAIL=False):
+    def _run_test(test):
+        test_name = test.__name__
+        try:
+            test()
+            print(("X" if XPASS else "") + f"PASS - {test_name}")
+        except Exception as e:
+            print(("X" if XFAIL else "") + f"FAIL - {test_name}")
+            print("Errors: ", e)
+        print(flush=True)
+
+    if len(args):
+        _run_test(args[0])
+    else:
+        return _run_test

python/test/lit.cfg.py

@@ -51,6 +51,9 @@ llvm_config.use_default_substitutions()
 # directories.
 config.excludes = ['lit.cfg.py', 'Inputs', 'Examples', 'CMakeLists.txt', 'README.txt', 'LICENSE.txt']
 
+if not bool(int(os.environ.get("TORCH_MLIR_ENABLE_LTC", 0))):
+   config.excludes.append("lazy_backend")
+
 # test_source_root: The root path where tests are located.
 config.test_source_root = os.path.dirname(__file__)
 

python/torch_mlir/csrc/base_lazy_backend/CMakeLists.txt

@@ -113,3 +113,25 @@ add_custom_command(
         COMMAND cp
                 ${PROJECT_SOURCE_DIR}/python/torch_mlir/csrc/base_lazy_backend/generated/*.h
                 ${TORCH_MLIR_PYTHON_PACKAGES_DIR}/torch_mlir/torch_mlir/base_lazy_backend/generated/)
+
+add_custom_command(
+        TARGET torch_mlir_ltc_backend POST_BUILD
+        COMMAND mkdir -p
+                ${TORCH_MLIR_PYTHON_PACKAGES_DIR}/torch_mlir/torch_mlir/base_lazy_backend/ops/)
+
+add_custom_command(
+        TARGET torch_mlir_ltc_backend POST_BUILD
+        COMMAND cp
+                ${PROJECT_SOURCE_DIR}/python/torch_mlir/csrc/base_lazy_backend/ops/*.h
+                ${TORCH_MLIR_PYTHON_PACKAGES_DIR}/torch_mlir/torch_mlir/base_lazy_backend/ops/)
+
+add_custom_command(
+        TARGET torch_mlir_ltc_backend POST_BUILD
+        COMMAND mkdir -p
+                ${TORCH_MLIR_PYTHON_PACKAGES_DIR}/torch_mlir/torch_mlir/base_lazy_backend/utils/)
+
+add_custom_command(
+        TARGET torch_mlir_ltc_backend POST_BUILD
+        COMMAND cp
+                ${PROJECT_SOURCE_DIR}/python/torch_mlir/csrc/base_lazy_backend/utils/*.h
+                ${TORCH_MLIR_PYTHON_PACKAGES_DIR}/torch_mlir/torch_mlir/base_lazy_backend/utils/)

python/torch_mlir/csrc/base_lazy_backend/backend_impl.h

@@ -15,6 +15,8 @@
 
 #pragma once
 
+#include <sstream>
+
 #include <torch/csrc/lazy/backend/backend_data.h>
 #include <torch/csrc/lazy/backend/backend_device.h>
 #include <torch/csrc/lazy/backend/backend_interface.h>
@@ -29,13 +31,20 @@ public:
     at::Tensor tensor;
     c10::optional<at::Scalar> scalar;
     bool requires_grad;
+    std::string name;
 
     Info() {}
     Info(const Info& other)
         : tensor{other.tensor}, scalar{other.scalar},
-          requires_grad{other.requires_grad} {}
+          requires_grad{other.requires_grad}, name{other.name} {}
     Info(const at::Tensor& tensor)
-        : tensor{tensor}, requires_grad{tensor.requires_grad()} {}
+        : tensor{tensor}, requires_grad{tensor.requires_grad()} {
+      static int num_tensors = 0;
+      std::ostringstream oss;
+      oss << "tensor" << num_tensors;
+      this->name = oss.str();
+      ++num_tensors;
+    }
     Info(const at::Scalar& scalar) : scalar{scalar}, requires_grad(false) {}
   };
 

python/torch_mlir/csrc/base_lazy_backend/mlir_lowering_context.h

@@ -135,11 +135,11 @@ public:
 
   MlirOperation func_op() const;
 
-  const std::string debug_string() const;
+  virtual const std::string debug_string() const;
 
-  const std::string to_string() const override;
+  virtual const std::string to_string() const override;
 
-private:
+protected:
   std::vector<std::string> parameter_names_;
   std::vector<Shape> parameter_shapes_;
   Shape result_shape_;

python/torch_mlir/csrc/base_lazy_backend/ops/device_data.cpp

@@ -3,6 +3,7 @@
 #include <torch/csrc/lazy/core/ir_builder.h>
 
 #include "device_data.h"
+#include "../backend_impl.h"
 
 namespace torch {
 namespace lazy {
@@ -13,11 +14,37 @@ DeviceData::DeviceData(std::shared_ptr<BackendData> data)
           data->shape(),
           /*num_outputs=*/1,
           /*hash_seed=*/static_cast<uint32_t>(101)),
-      data_(std::move(data)) {}
+      data_(std::move(data)) {
+  propagate_name();
+}
+
+void DeviceData::propagate_name() {
+  if (data_ && name_ != "") {
+    // Add device data name to backend data
+    TorchMlirBackendData* mlir_data = dynamic_cast<TorchMlirBackendData*>(data_.get());
+    TORCH_CHECK(mlir_data);
+    TorchMlirBackendData::Info* info = mlir_data->mlir_info();
+    TORCH_CHECK(info);
+    info->name = name_;
+  }
+}
+
+void DeviceData::SetData(std::shared_ptr<BackendData> data) {
+  data_ = data;
+  propagate_name();
+}
+
+void DeviceData::SetName(const std::string& name) {
+  name_ = name;
+  propagate_name();
+}
 
 std::string DeviceData::ToString() const {
   std::stringstream ss;
   ss << TorchMlirNode::ToString() << ", device=" << data_->device();
+  if (name_ != "") {
+     ss << ", name=" << name_;
+  }
   return ss.str();
 }
 

python/torch_mlir/csrc/base_lazy_backend/ops/device_data.h

@@ -27,13 +27,9 @@ class TORCH_API DeviceData : public TorchMlirNode {
 
   std::string ToString() const override;
 
-  const std::shared_ptr<BackendData>& data() const {
+  const std::shared_ptr<BackendData>& data() const { return data_; }
-    return data_;
-  }
 
-  void SetData(std::shared_ptr<BackendData> data) {
+  void SetData(std::shared_ptr<BackendData> data);
-    data_ = data;
-  }
 
   TorchMlirOpVector Lower(TorchMlirFunction function, TorchMlirLoweringContext* loctx) const override;
 
@@ -43,8 +39,14 @@ class TORCH_API DeviceData : public TorchMlirNode {
   // instead of calling the constructor directly.
   static NodePtr Create(std::shared_ptr<BackendData> data);
 
+  const std::string& GetName() const { return name_; }
+  void SetName(const std::string& name);
+
  private:
+  void propagate_name();
+
   std::shared_ptr<BackendData> data_;
+  std::string name_;
 };
 
 } // namespace lazy

python/torch_mlir/csrc/base_lazy_backend/utils/string_utils.h

@@ -0,0 +1,31 @@
+#pragma once
+
+#include <string>
+#include <sstream>
+#include <vector>
+
+
+template <typename T>
+std::ostream& string_join(std::ostream& out, const std::vector<T>& v, const std::string& delimiter) {
+    size_t i = 0;
+    for (const T& e : v) {
+        if ((i++) > 0) { out << delimiter; }
+        out << e;
+    }
+    return out;
+}
+
+template <typename T>
+std::string string_join(const std::vector<T>& v, const std::string& delimiter) {
+    std::ostringstream joined;
+    string_join(joined, v, delimiter);
+    return joined.str();
+}
+
+
+/*
+ * Returns true if str starts with prefix
+ */
+inline bool startswith(const std::string& str, const std::string& prefix) {
+   return str.rfind(prefix, 0) == 0;
+}

python/torch_mlir/csrc/base_lazy_backend/utils/tensor_utils.h

@@ -0,0 +1,30 @@
+#pragma once
+
+#include "torch/csrc/lazy/backend/backend_device.h"
+#include "torch/csrc/lazy/core/tensor.h"
+
+#include "../ops/device_data.h"
+
+
+namespace torch {
+namespace lazy {
+
+inline torch::lazy::DeviceData* device_data_cast(
+    const at::Tensor& tensor, c10::optional<torch::lazy::BackendDevice> device = c10::nullopt
+) {
+    if (!device) {
+        device = torch::lazy::GetBackendDevice(tensor);
+    }
+    TORCH_CHECK(device);
+    torch::lazy::LazyTensorPtr lazy_tensor = torch::lazy::GetLtcTensorOrCreateForWrappedNumber(tensor, *device);
+    if (lazy_tensor) {
+        torch::lazy::Value param_value = lazy_tensor->GetIrValue();
+        if (param_value && param_value->op() == torch::lazy::DeviceData::ClassOpKind()) {
+            return dynamic_cast<torch::lazy::DeviceData*>(param_value.node.get());
+        }
+    }
+    return nullptr;
+}
+
+}  // namespace lazy
+}  // namespace torch

python/torch_mlir/csrc/reference_lazy_backend/backend_impl.cpp

@@ -18,6 +18,7 @@
 #include <torch_mlir/csrc/base_lazy_backend/mlir_lowering_context.h>
 #include <torch_mlir/csrc/base_lazy_backend/utils/debug.h>
 #include <torch_mlir/csrc/base_lazy_backend/utils/exception.h>
+#include <torch_mlir/csrc/base_lazy_backend/utils/string_utils.h>
 
 #include "backend_impl.h"
 
@@ -88,6 +89,8 @@ public:
     auto mlir_computation =
         static_cast<TorchMlirComputation*>(computation.get());
 
+    int num_inputs = 0;
+
     // Vendor backend specific execution can be inserted here.
     //
     // We don't have a way to execute a computation based on the generated MLIR,
@@ -106,7 +109,17 @@ public:
         at::Tensor tensor = mlir_data->mlir_info()->tensor;
         stack.emplace_back(tensor);
       }
+
+      // count number of inputs
+      auto name = mlir_data->mlir_info()->name;
+      if (startswith(name, "input_")) {
+        // Printing tensor name for testing purposes
+        std::cout << "Input tensor: " << name << std::endl;
+        ++num_inputs;
+      }
     }
+    // Printing number of input tensors for testing purposes
+    std::cout << num_inputs << " input tensors found" << std::endl;
     graph_executor.run(stack);
     std::vector<torch::lazy::BackendDataPtr> results;
     for (torch::jit::IValue component : stack) {

python/torch_mlir/csrc/reference_lazy_backend/reference_lazy_backend_pybind.cpp

@@ -11,7 +11,9 @@
 #include "torch/csrc/lazy/backend/backend_interface.h"
 
 #include <torch_mlir/csrc/base_lazy_backend/mlir_lowering_context.h>
+#include <torch_mlir/csrc/base_lazy_backend/utils/string_utils.h>
 #include <torch_mlir/csrc/base_lazy_backend/utils/sys_utils.h>
+#include <torch_mlir/csrc/base_lazy_backend/utils/tensor_utils.h>
 
 #include <exception>
 #include <iostream>
@@ -73,6 +75,15 @@ PYBIND11_MODULE(_REFERENCE_LAZY_BACKEND, m) {
         torch::lazy::GetLatestComputation().get());
     return py::cast(computation);
   });
+  m.def("set_parameter_name",
+        [](const at::Tensor& tensor, const std::string& name) -> bool {
+            torch::lazy::DeviceData* ir_node = torch::lazy::device_data_cast(tensor);
+            if (ir_node) {
+                ir_node->SetName(name);
+                return true;
+            }
+            return false;
+        });
   m.def("_initialize", []() {
     NoGilSection gil;
     Initialize();

setup.py

@@ -20,6 +20,10 @@
 # prevent this script from attempting to build the directory, and will simply
 # use the (presumed already built) directory as-is.
 #
+# By default the lazy tensor backend is disabled and not built to avoid conflicts
+# with the out-of-tree build. To enable it, set the TORCH_MLIR_ENABLE_LTC
+# environment variable to 1.
+#
 # The package version can be set with the TORCH_MLIR_PYTHON_PACKAGE_VERSION
 # environment variable. For example, this can be "20220330.357" for a snapshot
 # release on 2022-03-30 with build number 357.
@@ -64,7 +68,6 @@ class CMakeBuild(build_py):
         python_package_dir = os.path.join(cmake_build_dir,
                                           "tools", "torch-mlir", "python_packages",
                                           "torch_mlir")
-
         if not os.getenv("TORCH_MLIR_CMAKE_BUILD_DIR_ALREADY_BUILT"):
             src_dir = os.path.abspath(os.path.dirname(__file__))
             llvm_dir = os.path.join(
@@ -83,6 +86,11 @@ class CMakeBuild(build_py):
                 f"-DCMAKE_C_VISIBILITY_PRESET=hidden",
                 f"-DCMAKE_CXX_VISIBILITY_PRESET=hidden",
             ]
+            # TODO: Enable LTC by default once JIT importer linkage issue is fixed (https://github.com/llvm/torch-mlir/issues/1154)
+            enable_ltc = bool(int(os.environ.get("TORCH_MLIR_ENABLE_LTC", 0)))
+            if not enable_ltc:
+                cmake_args.append("-DTORCH_MLIR_ENABLE_LTC=OFF")
+
             os.makedirs(cmake_build_dir, exist_ok=True)
             cmake_cache_file = os.path.join(cmake_build_dir, "CMakeCache.txt")
             if os.path.exists(cmake_cache_file):