Remove caffe2 from more build files (pytorch#125898)

Co-authored-by: Aaron Gokaslan <[email protected]>
Pull Request resolved: pytorch#125898
Approved by: https://github.com/Skylion007

.ci/pytorch/build.sh

@@ -44,11 +44,6 @@ if [[ "$BUILD_ENVIRONMENT" == *cuda11* ]]; then
   fi
 fi
 
-if [[ ${BUILD_ENVIRONMENT} == *"caffe2"* ]]; then
-  echo "Caffe2 build is ON"
-  export BUILD_CAFFE2=ON
-fi
-
 if [[ ${BUILD_ENVIRONMENT} == *"paralleltbb"* ]]; then
   export ATEN_THREADING=TBB
   export USE_TBB=1

CMakeLists.txt

@@ -181,11 +181,7 @@ option(BUILD_BINARY "Build C++ binaries" OFF)
 option(BUILD_DOCS "Build Caffe2 documentation" OFF)
 option(BUILD_CUSTOM_PROTOBUF "Build and use Caffe2's own protobuf under third_party" ON)
 option(BUILD_PYTHON "Build Python binaries" ON)
-option(BUILD_CAFFE2 "Master flag to build Caffe2" OFF)
 option(BUILD_LITE_INTERPRETER "Master flag to build Lite Interpreter" OFF)
-cmake_dependent_option(
-    BUILD_CAFFE2_OPS "Build Caffe2 operators" ON
-    "BUILD_CAFFE2" OFF)
 option(BUILD_SHARED_LIBS "Build libcaffe2.so" ON)
 cmake_dependent_option(
     CAFFE2_LINK_LOCAL_PROTOBUF "If set, build protobuf inside libcaffe2.so." ON
@@ -635,7 +631,6 @@ if(INTERN_BUILD_MOBILE)
   endif()
   set(BUILD_PYTHON OFF)
   set(BUILD_FUNCTORCH OFF)
-  set(BUILD_CAFFE2_OPS OFF)
   set(USE_DISTRIBUTED OFF)
   set(NO_API ON)
   set(USE_FBGEMM OFF)
@@ -1208,13 +1203,6 @@ else()
       "shared libs.")
 endif()
 
-# ---[ Modules
-# If master flag for buildling Caffe2 is disabled, we also disable the
-# build for Caffe2 related operator modules.
-if(BUILD_CAFFE2)
-  add_subdirectory(modules)
-endif()
-
 # ---[ Binaries
 # Binaries will be built after the Caffe2 main libraries and the modules
 # are built. For the binaries, they will be linked to the Caffe2 main

CONTRIBUTING.md

@@ -667,7 +667,6 @@ only interested in a specific component.
 - Working on a test binary? Run `(cd build && ninja bin/test_binary_name)` to
   rebuild only that test binary (without rerunning cmake). (Replace `ninja` with
   `make` if you don't have ninja installed).
-- Don't need Caffe2?  Pass `BUILD_CAFFE2=0` to disable Caffe2 build.
 
 On the initial build, you can also speed things up with the environment
 variables `DEBUG`, `USE_DISTRIBUTED`, `USE_MKLDNN`, `USE_CUDA`, `USE_FLASH_ATTENTION`, `USE_MEM_EFF_ATTENTION`, `BUILD_TEST`, `USE_FBGEMM`, `USE_NNPACK` and `USE_QNNPACK`.
@@ -1196,7 +1195,7 @@ build_with_asan()
   LDFLAGS="-stdlib=libstdc++" \
   CFLAGS="-fsanitize=address -fno-sanitize-recover=all -shared-libasan -pthread" \
   CXX_FLAGS="-pthread" \
-  USE_CUDA=0 USE_OPENMP=0 BUILD_CAFFE2_OPS=0 USE_DISTRIBUTED=0 DEBUG=1 \
+  USE_CUDA=0 USE_OPENMP=0 USE_DISTRIBUTED=0 DEBUG=1 \
   python setup.py develop
 }
 

README.md

@@ -379,7 +379,7 @@ You can also pass the `CMAKE_VARS="..."` environment variable to specify additio
 See [setup.py](./setup.py) for the list of available variables.
 
 ```bash
-CMAKE_VARS="BUILD_CAFFE2=ON BUILD_CAFFE2_OPS=ON" make -f docker.Makefile
+make -f docker.Makefile
 ```
 
 ### Building the Documentation

aten/src/ATen/CMakeLists.txt

@@ -54,7 +54,7 @@ if(NOT BUILD_LITE_INTERPRETER)
 endif()
 EXCLUDE(ATen_CORE_SRCS "${ATen_CORE_SRCS}" ${ATen_CORE_TEST_SRCS})
 # Exclude TensorImpl_test.cpp if compiling without Caffe2
-if(NOT BUILD_CAFFE2 AND NOT BUILD_LITE_INTERPRETER)
+if(NOT BUILD_LITE_INTERPRETER)
   file(GLOB_RECURSE ATen_CORE_EXCLUDED_TEST_SRCS "core/TensorImpl_test.cpp")
   EXCLUDE(ATen_CORE_TEST_SRCS "${ATen_CORE_TEST_SRCS}" ${ATen_CORE_EXCLUDED_TEST_SRCS})
 endif()

aten/src/ATen/test/CMakeLists.txt

@@ -123,16 +123,6 @@ list(APPEND ATen_XPU_TEST_SRCS
   ${CMAKE_CURRENT_SOURCE_DIR}/xpu_generator_test.cpp
   )
 
-# Caffe2 specific tests
-if(BUILD_CAFFE2)
-  list(APPEND ATen_CPU_TEST_SRCS
-    ${CMAKE_CURRENT_SOURCE_DIR}/ExclusivelyOwned_test.cpp
-    ${CMAKE_CURRENT_SOURCE_DIR}/tensor_interop_test.cpp)
-  list(APPEND ATen_CUDA_TEST_SRCS
-    ${CMAKE_CURRENT_SOURCE_DIR}/cuda_tensor_interop_test.cpp)
-endif()
-
-
 # ---[ Send the lists to the parent scope.
 set(ATen_CPU_TEST_SRCS ${ATen_CPU_TEST_SRCS} PARENT_SCOPE)
 set(ATen_CUDA_TEST_SRCS ${ATen_CUDA_TEST_SRCS} PARENT_SCOPE)

binaries/CMakeLists.txt

@@ -7,16 +7,6 @@ if(INTERN_BUILD_MOBILE)
   return()
 endif()
 
-if(BUILD_CAFFE2)
-  caffe2_binary_target("at_launch_benchmark.cc")
-  target_include_directories(at_launch_benchmark PUBLIC
-    ${CMAKE_BINARY_DIR}/aten/src)
-
-  caffe2_binary_target("intra_inter_benchmark.cc")
-  target_include_directories(intra_inter_benchmark PUBLIC
-    ${CMAKE_BINARY_DIR}/aten/src)
-endif()
-
 caffe2_binary_target("parallel_info.cc")
 target_include_directories(parallel_info PUBLIC
   ${CMAKE_BINARY_DIR}/aten/src) # provides "ATen/TypeExtendedInterface.h" to ATen.h

buckbuild.bzl

@@ -279,7 +279,6 @@ def get_pt_preprocessor_flags():
         "-D_THP_CORE",
         "-DUSE_SCALARS",
         "-DNO_CUDNN_DESTROY_HANDLE",
-        "-DBUILD_CAFFE2",
     ]
 
     if _is_build_mode_dev():

caffe2/CMakeLists.txt

@@ -110,21 +110,11 @@ endif()
 add_subdirectory(core)
 add_subdirectory(serialize)
 add_subdirectory(utils)
-if(BUILD_CAFFE2 OR (NOT USE_FBGEMM))
+if(NOT USE_FBGEMM)
   add_subdirectory(perfkernels)
 endif()
 
-# Skip modules that are not used by libtorch mobile yet.
-if(BUILD_CAFFE2 AND NOT INTERN_BUILD_MOBILE)
-  add_subdirectory(core/nomnigraph)
-  if(USE_NVRTC)
-    add_subdirectory(cuda_rtc)
-  endif()
-  if(BUILD_CAFFE2_OPS)
-  endif()
-  add_subdirectory(proto)
-endif()
-if(NOT BUILD_CAFFE2 AND NOT INTERN_BUILD_MOBILE)
+if(NOT INTERN_BUILD_MOBILE)
   add_subdirectory(proto)
 endif()
 
@@ -585,17 +575,10 @@ if(NOT INTERN_BUILD_MOBILE AND NOT BUILD_LITE_INTERPRETER)
     ${TORCH_SRC_DIR}/csrc/utils/byte_order.cpp
   )
 
-  # Disable legacy import of building without Caffe2 support
-  if(BUILD_CAFFE2)
-    list(APPEND TORCH_SRCS
-      ${TORCH_SRC_DIR}/csrc/jit/serialization/import_legacy.cpp
-    )
-  else()
-    set_source_files_properties(
-      ${TORCH_SRC_DIR}/csrc/jit/serialization/import.cpp
-      PROPERTIES COMPILE_FLAGS "-DC10_DISABLE_LEGACY_IMPORT"
-    )
-  endif()
+  set_source_files_properties(
+    ${TORCH_SRC_DIR}/csrc/jit/serialization/import.cpp
+    PROPERTIES COMPILE_FLAGS "-DC10_DISABLE_LEGACY_IMPORT"
+  )
   if(USE_DISTRIBUTED)
     append_filelist("libtorch_distributed_base_sources" TORCH_SRCS)
     if(NOT WIN32)
@@ -809,11 +792,6 @@ if(HAVE_SOVERSION)
       VERSION ${TORCH_VERSION} SOVERSION ${TORCH_SOVERSION})
 endif()
 torch_compile_options(torch_cpu)  # see cmake/public/utils.cmake
-if(BUILD_CAFFE2 AND NOT MSVC)
-  # Caffe2 has too many signed-unsigned violation, but the framework is dead
-  # So no point in fixing those
-  target_compile_options(torch_cpu PRIVATE "-Wno-sign-compare")
-endif()
 
 # Ignore Wdeprecated-XXX errors from third-party libraries
 if(NOT MSVC)
@@ -1921,14 +1899,6 @@ if(BUILD_TEST)
       endif()
     endforeach()
   endif()
-
-  # For special tests that explicitly uses dependencies, we add them here
-  if(BUILD_CAFFE2 AND USE_MPI)
-    target_link_libraries(mpi_test MPI::MPI_CXX)
-    if(USE_CUDA)
-      target_link_libraries(mpi_gpu_test MPI::MPI_CXX)
-    endif()
-  endif()
 endif()
 
 if(MSVC)
@@ -1998,11 +1968,6 @@ if(BUILD_PYTHON)
     set_source_files_properties(${TORCH_SRC_DIR}/../caffe2/operators/box_with_nms_limit_op.cc PROPERTIES COMPILE_FLAGS -Wno-attributes)
   endif()
 
-  # ---[ Python.
-  if(BUILD_CAFFE2)
-  target_compile_definitions(torch PRIVATE BUILD_CAFFE2)
-  endif()
-
   # generated pb files are copied from build/caffe2 to caffe2
   # if we copied them back to build this would create a build cycle
   # consider removing the need for globs

caffe2/__init__.py

@@ -2,5 +2,4 @@
 from torch.onnx import _CAFFE2_ATEN_FALLBACK
 
 if not _CAFFE2_ATEN_FALLBACK:
-    warnings.warn("Caffe2 support is not fully enabled in this PyTorch build. "
-                  "Please enable Caffe2 by building PyTorch from source with `BUILD_CAFFE2=1` flag.")
+    warnings.warn("Caffe2 support is no longer present in PyTorch.")

caffe2/core/CMakeLists.txt

@@ -1,68 +1,4 @@
-if(NOT BUILD_CAFFE2 OR INTERN_BUILD_MOBILE)
-  list(APPEND Caffe2_CPU_SRCS
-    "${CMAKE_CURRENT_SOURCE_DIR}/common.cc"
-  )
-  set(Caffe2_CPU_SRCS ${Caffe2_CPU_SRCS} PARENT_SCOPE)
-  return()
-endif()
-
-# ---[ GPU files
-# ------[ cuDNN
-if(USE_CUDNN)
-  file(GLOB tmp *_cudnn.cc)
-  set(Caffe2_GPU_SRCS ${Caffe2_GPU_SRCS} ${tmp})
-endif()
-# ------[ general GPU
-file(GLOB tmp *_gpu.cc)
-set(Caffe2_GPU_SRCS ${Caffe2_GPU_SRCS} ${tmp})
-# ------[ CUDA sources
-file(GLOB tmp *.cu)
-set(Caffe2_GPU_SRCS ${Caffe2_GPU_SRCS} ${tmp})
-# exclude test files
-file(GLOB tmp *_test.cc)
-exclude(Caffe2_GPU_SRCS "${Caffe2_GPU_SRCS}" ${tmp})
-
-# ---[ general HIP files
-file(GLOB tmp hip/*.cc)
-set(Caffe2_HIP_SRCS ${Caffe2_HIP_SRCS} ${tmp})
-# ------[ HIP sources
-file(GLOB tmp hip/*.hip)
-set(Caffe2_HIP_SRCS ${Caffe2_HIP_SRCS} ${tmp})
-# exclude test files
-file(GLOB tmp hip/*_test.cc)
-exclude(Caffe2_HIP_SRCS "${Caffe2_HIP_SRCS}" ${tmp})
-
-# ---[ CPU files.
-file(GLOB tmp *.cc)
-# Manually remove the cudnn files since we might be using USE_CUDNN=OFF
-# TODO: when we move to explicit file list, this would not be needed.
-file(GLOB tmp_cudnn *_cudnn.cc)
-exclude(tmp "${tmp}" ${tmp_cudnn})
-set(Caffe2_CPU_SRCS ${Caffe2_CPU_SRCS} ${tmp})
-# exclude test files and gpu files
-file(GLOB tmp *_test.cc)
-exclude(Caffe2_CPU_SRCS "${Caffe2_CPU_SRCS}" ${tmp})
-exclude(Caffe2_CPU_SRCS "${Caffe2_CPU_SRCS}" ${Caffe2_GPU_SRCS})
-exclude(Caffe2_CPU_SRCS "${Caffe2_CPU_SRCS}" ${Caffe2_HIP_SRCS})
-
-# ---[ GPU test files
-file(GLOB tmp *_gpu_test.cc)
-set(Caffe2_GPU_TEST_SRCS ${Caffe2_GPU_TEST_SRCS} ${tmp})
-
-# ---[ HIP test files
-file(GLOB tmp hip/*_test.cc)
-set(Caffe2_HIP_TEST_SRCS ${Caffe2_HIP_TEST_SRCS} ${tmp})
-
-# ---[ CPU test files
-file(GLOB tmp *_test.cc)
-set(Caffe2_CPU_TEST_SRCS ${Caffe2_CPU_TEST_SRCS} ${tmp})
-exclude(Caffe2_CPU_TEST_SRCS "${Caffe2_CPU_TEST_SRCS}" ${Caffe2_GPU_TEST_SRCS})
-exclude(Caffe2_CPU_TEST_SRCS "${Caffe2_CPU_TEST_SRCS}" ${Caffe2_HIP_TEST_SRCS})
-
-# ---[ Send the lists to the parent scope.
+list(APPEND Caffe2_CPU_SRCS
+  "${CMAKE_CURRENT_SOURCE_DIR}/common.cc"
+)
 set(Caffe2_CPU_SRCS ${Caffe2_CPU_SRCS} PARENT_SCOPE)
-set(Caffe2_GPU_SRCS ${Caffe2_GPU_SRCS} PARENT_SCOPE)
-set(Caffe2_HIP_SRCS ${Caffe2_HIP_SRCS} PARENT_SCOPE)
-set(Caffe2_CPU_TEST_SRCS ${Caffe2_CPU_TEST_SRCS} PARENT_SCOPE)
-set(Caffe2_GPU_TEST_SRCS ${Caffe2_GPU_TEST_SRCS} PARENT_SCOPE)
-set(Caffe2_HIP_TEST_SRCS ${Caffe2_HIP_TEST_SRCS} PARENT_SCOPE)

caffe2/core/blob_serialization.cc

@@ -1,5 +1,6 @@
 #include "caffe2/core/blob_serialization.h"
 
+#include <limits>
 #include <mutex>
 #include <sstream>
 #include <utility>
@@ -83,8 +84,7 @@ Range<T*> GetMutableTensorDataRange(
     size_t start,
     size_t numElements) {
   CAFFE_ENFORCE(
-      // NOLINTNEXTLINE(clang-diagnostic-sign-compare)
-      start + numElements <= tensor.numel(),
+      static_cast<int64_t>(start + numElements) <= tensor.numel(),
       "Requested invalid mutable tensor range [",
       start,
       ", ",
@@ -100,8 +100,7 @@ c10::ArrayRef<T> GetTensorDataRange(
     size_t start,
     size_t numElements) {
   CAFFE_ENFORCE(
-      // NOLINTNEXTLINE(clang-diagnostic-sign-compare)
-      start + numElements <= tensor.numel(),
+      static_cast<int64_t>(start + numElements) <= tensor.numel(),
       "Requested invalid tensor range [",
       start,
       ", ",
@@ -390,8 +389,7 @@ void TensorSerializer::SerializeWithOptions(
   // Poorman's IOBound ThreadPool
   SimpleQueue<size_t> chunkQueue;
   auto task = [&]() {
-    // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-    size_t chunkStart;
+    size_t chunkStart = std::numeric_limits<size_t>::max();
     while (chunkQueue.Pop(&chunkStart)) {
       processChunk(chunkStart);
     }
@@ -409,8 +407,7 @@ void TensorSerializer::SerializeWithOptions(
   VLOG(1) << "Serializing blob " << name;
   // Serialize whole vector. If vector is empty, it's shape still needs to be
   // serialized in empty proto
-  for (size_t chunkBegin = 0;
-       // NOLINTNEXTLINE(clang-diagnostic-sign-compare)
+  for (int64_t chunkBegin = 0;
        chunkBegin < std::max(tensor.numel(), static_cast<int64_t>(1));
        chunkBegin += chunk_size) {
     VLOG(2) << "Starting a chunk at " << chunkBegin;
@@ -582,8 +579,7 @@ void SerializeTensorData(const SerializeParams<float>& params) {
       BlobSerializationOptions_FloatFormat_FLOAT_BFLOAT16) {
     // NOLINTNEXTLINE(cppcoreguidelines-avoid-c-arrays,modernize-avoid-c-arrays)
     std::unique_ptr<float[]> tmp_buffer;
-    // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-    const float* src;
+    const float* src = nullptr;
     if (params.context.device() == CPU) {
       src = params.input.data();
     } else {
@@ -653,14 +649,12 @@ void TensorSerializer::Serialize(
     size_t chunkBegin,
     int32_t chunkSize) {
   CAFFE_ENFORCE(
-      // NOLINTNEXTLINE(clang-diagnostic-sign-compare)
-      chunkBegin <= input.numel(),
+      static_cast<int64_t>(chunkBegin) <= input.numel(),
       "Chunk begin is out of tensor: ",
       chunkBegin,
       ' ',
       input.numel());
-  // NOLINTNEXTLINE(clang-diagnostic-sign-compare)
-  if (chunkBegin + chunkSize > input.numel()) {
+  if (static_cast<int64_t>(chunkBegin + chunkSize) > input.numel()) {
     chunkSize = input.numel() - chunkBegin;
   }
 
@@ -1029,8 +1023,7 @@ DESERIALIZE_IMPL(float, FMT_BFLOAT16) {
       params.tensor_proto.raw_data().data());
 
   // If we are on a big-endian machine, byte-swap the serialized data.
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-  const fbgemm::bfloat16* src;
+  const fbgemm::bfloat16* src = nullptr;
   // NOLINTNEXTLINE(cppcoreguidelines-avoid-c-arrays,modernize-avoid-c-arrays)
   std::unique_ptr<fbgemm::bfloat16[]> bswap_buffer;
   if (kIsLittleEndian) {
@@ -1045,8 +1038,7 @@ DESERIALIZE_IMPL(float, FMT_BFLOAT16) {
   // bfloat16 to float conversion.
   // NOLINTNEXTLINE(cppcoreguidelines-avoid-c-arrays,modernize-avoid-c-arrays)
   std::unique_ptr<float[]> tmp_buffer;
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-  float* dest;
+  float* dest = nullptr;
   if (params.context.device() == CPU) {
     dest = params.dest.data();
   } else {

caffe2/proto/CMakeLists.txt

@@ -1,8 +1,4 @@
-if(BUILD_CAFFE2)
-  file(GLOB Caffe2_PROTOBUF_FILES "${CMAKE_CURRENT_SOURCE_DIR}/*.proto")
-else()
-  set(Caffe2_PROTOBUF_FILES "${CMAKE_CURRENT_SOURCE_DIR}/torch.proto;${CMAKE_CURRENT_SOURCE_DIR}/caffe2.proto")
-endif()
+set(Caffe2_PROTOBUF_FILES "${CMAKE_CURRENT_SOURCE_DIR}/torch.proto;${CMAKE_CURRENT_SOURCE_DIR}/caffe2.proto")
 
 caffe2_protobuf_generate_cpp_py(Caffe2_PROTO_SRCS Caffe2_PROTO_HEADERS Caffe2_PROTO_PY ${Caffe2_PROTOBUF_FILES})