Fix TorchRec inference solution with updated torch.fx + TorchScript Solution (#2101)

Summary:
Previously, the TorchRec inference solution used torch.deploy, which isn't supported, and the solution itself was outdated and broken. This PR revamps the inference example with Torch.FX and TorchScript, which more accurately represents what is used in production for RecSys currently. Furthermore, the example was constructed with simplicity as the top priority, with fewer package requirements and steps to run. This is the first iteration of the inference example, with more to come!

Pull Request resolved: #2101

Differential Revision: D58478461

Pulled By: PaulZhang12

Author: PaulZhang12

examples/inference/README.md renamed to examples/inference_legacy/README.md

@@ -1,113 +1,76 @@
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# All rights reserved.
+# Copyright 2018 gRPC authors.
 #
-# This source code is licensed under the BSD-style license found in the
-# LICENSE file in the root directory of this source tree.
-
-cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
-project(inference)
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+# cmake build file for C++ helloworld example.
+# Assumes protobuf and gRPC have been installed using cmake.
+# See cmake_externalproject/CMakeLists.txt for all-in-one cmake build
+# that automatically builds all the dependencies before building helloworld.
 
-# This step is crucial to ensure that the
-# _REFLECTION, _GRPC_GRPCPP and _PROTOBUF_LIBPROTOBUF variables are set.
-# e.g. ~/gprc/examples/cpp/cmake/common.cmake
-include(${GRPC_COMMON_CMAKE_PATH}/common.cmake)
+cmake_minimum_required(VERSION 3.8)
 
+project(inference C CXX)
 
-# abi and other flags
+include(/home/paulzhan/grpc/examples/cpp/cmake/common.cmake)
 
-if(DEFINED GLIBCXX_USE_CXX11_ABI)
-  if(${GLIBCXX_USE_CXX11_ABI} EQUAL 1)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -D_GLIBCXX_USE_CXX11_ABI=1")
-    set(TORCH_CXX_FLAGS "-D_GLIBCXX_USE_CXX11_ABI=1")
-  endif()
-endif()
 
-# keep it static for now since folly-shared version is broken
-# set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC")
-# set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fPIC")
+# Proto file
+get_filename_component(hw_proto "/home/paulzhan/torchrec/torchrec/inference/protos/predictor.proto" ABSOLUTE)
+get_filename_component(hw_proto_path "${hw_proto}" PATH)
 
-# dependencies
-find_package(Boost REQUIRED)
-find_package(Torch REQUIRED)
-find_package(folly REQUIRED)
-find_package(gflags REQUIRED)
+# Generated sources
+set(hw_proto_srcs "${CMAKE_CURRENT_BINARY_DIR}/predictor.pb.cc")
+set(hw_proto_hdrs "${CMAKE_CURRENT_BINARY_DIR}/predictor.pb.h")
+set(hw_grpc_srcs "${CMAKE_CURRENT_BINARY_DIR}/predictor.grpc.pb.cc")
+set(hw_grpc_hdrs "${CMAKE_CURRENT_BINARY_DIR}/predictor.grpc.pb.h")
+add_custom_command(
+      OUTPUT "${hw_proto_srcs}" "${hw_proto_hdrs}" "${hw_grpc_srcs}" "${hw_grpc_hdrs}"
+      COMMAND ${_PROTOBUF_PROTOC}
+      ARGS --grpc_out "${CMAKE_CURRENT_BINARY_DIR}"
+        --cpp_out "${CMAKE_CURRENT_BINARY_DIR}"
+        -I "${hw_proto_path}"
+        --plugin=protoc-gen-grpc="${_GRPC_CPP_PLUGIN_EXECUTABLE}"
+        "${hw_proto}"
+      DEPENDS "${hw_proto}")
 
-include_directories(${Torch_INCLUDE_DIRS})
-include_directories(${folly_INCLUDE_DIRS})
-include_directories(${PYTORCH_FMT_INCLUDE_PATH})
+# Include generated *.pb.h files
+include_directories("${CMAKE_CURRENT_BINARY_DIR}")
 
 set(CMAKE_CXX_STANDARD 17)
 
-# torch deploy library
-add_library(torch_deploy_internal STATIC
-    ${DEPLOY_INTERPRETER_PATH}/libtorch_deployinterpreter.o
-    ${DEPLOY_SRC_PATH}/deploy.cpp
-    ${DEPLOY_SRC_PATH}/loader.cpp
-    ${DEPLOY_SRC_PATH}/path_environment.cpp
-    ${DEPLOY_SRC_PATH}/elf_file.cpp)
-
-# For python builtins. caffe2_interface_library properly
-# makes use of the --whole-archive option.
-target_link_libraries(torch_deploy_internal PRIVATE
-  crypt pthread dl util m z ffi lzma readline nsl ncursesw panelw
-)
-target_link_libraries(torch_deploy_internal
-  PUBLIC shm torch ${PYTORCH_LIB_FMT}
-)
-caffe2_interface_library(torch_deploy_internal torch_deploy)
-
-# inference library
-
-# for our own header files
-include_directories(include/)
-include_directories(gen/)
-
-# define our library target
-add_library(inference STATIC
-  src/Batching.cpp
-  src/BatchingQueue.cpp
-  src/GPUExecutor.cpp
-  src/ResultSplit.cpp
-  src/Exception.cpp
-  src/ResourceManager.cpp
-)
-
-# -rdynamic is needed to link against the static library
-target_link_libraries(inference "-Wl,--no-as-needed -rdynamic"
-  dl torch_deploy "${TORCH_LIBRARIES}" ${FBGEMM_LIB} ${FOLLY_LIBRARIES}
-)
-
-# for generated protobuf
-
-# grpc headers. e.g. ~/.local/include
-include_directories(${GRPC_HEADER_INCLUDE_PATH})
-
-set(pred_grpc_srcs "gen/torchrec/inference/predictor.grpc.pb.cc")
-set(pred_grpc_hdrs "gen/torchrec/inference/predictor.grpc.pb.h")
-set(pred_proto_srcs "gen/torchrec/inference/predictor.pb.cc")
-set(pred_proto_hdrs "gen/torchrec/inference/predictor.pb.h")
+# Torch + FBGEMM
+find_package(Torch REQUIRED)
+add_library( fbgemm SHARED IMPORTED GLOBAL )
+set_target_properties(fbgemm PROPERTIES IMPORTED_LOCATION ${FBGEMM_LIB})
 
-add_library(pred_grpc_proto STATIC
-  ${pred_grpc_srcs}
-  ${pred_grpc_hdrs}
-  ${pred_proto_srcs}
-  ${pred_proto_hdrs})
 
-target_link_libraries(pred_grpc_proto
+add_library(hw_grpc_proto STATIC
+  ${hw_grpc_srcs}
+  ${hw_grpc_hdrs}
+  ${hw_proto_srcs}
+  ${hw_proto_hdrs})
+target_link_libraries(hw_grpc_proto
   ${_REFLECTION}
   ${_GRPC_GRPCPP}
   ${_PROTOBUF_LIBPROTOBUF})
 
-# server
+# Targets greeter_[async_](client|server)
 add_executable(server server.cpp)
 target_link_libraries(server
-  inference
-  torch_deploy
-  pred_grpc_proto
   "${TORCH_LIBRARIES}"
-  ${FOLLY_LIBRARIES}
-  ${PYTORCH_LIB_FMT}
-  ${FBGEMM_LIB}
+  fbgemm
+  hw_grpc_proto
   ${_REFLECTION}
   ${_GRPC_GRPCPP}
-  ${_PROTOBUF_LIBPROTOBUF})
+  ${_PROTOBUF_LIBPROTOBUF}
+)