Move flags from file generation to compile definition (#4173)

* port with cuda

* refactor generated_compile_flags with cuda

* use_cxx11_abi

* refactor

* larger tol

Co-authored-by: Tsai <caishenghang@oneflow.org>
Co-authored-by: oneflow-ci-bot <69100618+oneflow-ci-bot@users.noreply.github.com>

cmake/oneflow.cmake

@@ -262,28 +262,8 @@ add_custom_target(of_pyscript_copy ALL
     COMMAND ${CMAKE_COMMAND} -E touch "${of_pyscript_dir}/oneflow/core/__init__.py"
     COMMAND ${CMAKE_COMMAND} -E make_directory "${of_pyscript_dir}/oneflow/python_gen"
     COMMAND ${CMAKE_COMMAND} -E touch "${of_pyscript_dir}/oneflow/python_gen/__init__.py"
-    COMMAND echo "generated_compile_flags = []" > "${of_pyscript_dir}/oneflow/python_gen/sysconfig.py"
     COMMAND ${Python_EXECUTABLE} "${PROJECT_SOURCE_DIR}/tools/generate_oneflow_symbols_export_file.py"
         "${PROJECT_SOURCE_DIR}" "${of_pyscript_dir}/oneflow/python_gen/__export_symbols__.py")
-if (BUILD_CUDA)
-  add_custom_command(TARGET of_pyscript_copy POST_BUILD
-        COMMAND echo "with_cuda=True" >> "${of_pyscript_dir}/oneflow/python_gen/compatibility.py"
-        COMMAND echo "\"generated_compile_flags.append('-DWITH_CUDA')\"" >> ${of_pyscript_dir}/oneflow/python_gen/sysconfig.py
-        )
-else()
-  add_custom_command(TARGET of_pyscript_copy POST_BUILD
-        COMMAND echo "with_cuda=False" >> "${of_pyscript_dir}/oneflow/python_gen/compatibility.py")
-endif()
-
-if (USE_CXX11_ABI)
-  add_custom_command(TARGET of_pyscript_copy POST_BUILD
-        COMMAND echo "\"generated_compile_flags.append('-D_GLIBCXX_USE_CXX11_ABI=1')\"" >> ${of_pyscript_dir}/oneflow/python_gen/sysconfig.py
-        )
-else()
-  add_custom_command(TARGET of_pyscript_copy POST_BUILD
-        COMMAND echo "\"generated_compile_flags.append('-D_GLIBCXX_USE_CXX11_ABI=0')\"" >> ${of_pyscript_dir}/oneflow/python_gen/sysconfig.py
-        )
-endif()
 
 add_dependencies(of_pyscript_copy of_protoobj)
 add_custom_target(generate_api ALL

oneflow/api/python/flags.cpp

+/*
+Copyright 2020 The OneFlow Authors. All rights reserved.
+
+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.
+*/
+#include "oneflow/api/python/of_api_registry.h"
+
+namespace oneflow {
+
+ONEFLOW_API_PYBIND11_MODULE("flags", m) {
+  m.def("with_cuda", []() {
+#ifdef WITH_CUDA
+    return true;
+#else
+    return false;
+#endif  // WITH_CUDA
+  });
+
+  m.def("use_cxx11_abi", []() {
+#if _GLIBCXX_USE_CXX11_ABI == 1
+    return true;
+#else
+    return false;
+#endif  // _GLIBCXX_USE_CXX11_ABI
+  });
+}
+
+}  // namespace oneflow

oneflow/python/framework/config_util.py

@@ -81,16 +81,14 @@ def api_gpu_device_num(val: int) -> None:
         val (int): number of GPUs. It is identical on every machine. In other words,
         you can't specify different number of GPUs you would like to use on each machine.
     """
-    from oneflow.python_gen.compatibility import with_cuda
-
-    if with_cuda == False:
+    if oneflow_api.flags.with_cuda():
+        return enable_if.unique([gpu_device_num, do_nothing])(val)
+    else:
         print(
             "INFO: for CPU-only OneFlow, oneflow.config.gpu_device_num is equivalent to oneflow.config.cpu_device_num"
         )
         print(traceback.format_stack()[-2])
         return enable_if.unique([cpu_device_num, do_nothing])(val)
-    else:
-        return enable_if.unique([gpu_device_num, do_nothing])(val)
 
 
 @enable_if.condition(hob.in_normal_mode & ~hob.session_initialized)

oneflow/python/framework/placement_util.py

@@ -23,6 +23,7 @@ from oneflow.python.oneflow_export import oneflow_export, oneflow_deprecate
 import oneflow.python.lib.core.enable_if as enable_if
 import oneflow
 import traceback
+import oneflow_api
 
 
 @oneflow_export("device_prior_placement", "fixed_placement")
@@ -54,18 +55,18 @@ def api_placement(
 
     For example:
 
-    If you run program on single machine, you can assign the specified device like this: 
+    If you run program on single machine, you can assign the specified device like this:
 
-    .. code-block:: python 
+    .. code-block:: python
 
         with flow.scope.placement("gpu", "0:0"):
             logits = lenet(images, train=False)
             loss = flow.nn.sparse_softmax_cross_entropy_with_logits(labels, logits, name="softmax_loss")
             flow.losses.add_loss(loss)
 
-    Or you run distributed program, you can assign the specified devices like this: 
+    Or you run distributed program, you can assign the specified devices like this:
 
-    .. code-block:: python 
+    .. code-block:: python
 
         # configure machines ids, ips, etc.
         with flow.scope.placement("gpu", ['0:0-7', '1:0-7']):
@@ -74,9 +75,8 @@ def api_placement(
             flow.losses.add_loss(loss)
 
     """
-    from oneflow.python_gen.compatibility import with_cuda
 
-    if with_cuda == False:
+    if oneflow_api.flags.with_cuda() == False:
         device_tag = "cpu"
     func = enable_if.unique(
         [

oneflow/python/framework/session_util.py

@@ -489,11 +489,9 @@ def _TryCompleteConfigProto(config_proto):
 
 
 def _GetDefaultConfigProto():
-    from oneflow.python_gen.compatibility import with_cuda
-
     config_proto = job_set_util.ConfigProto()
     config_proto.resource.machine_num = 0
-    if with_cuda:
+    if oneflow_api.flags.with_cuda():
         config_proto.resource.gpu_device_num = 1
     else:
         config_proto.resource.cpu_device_num = 1

oneflow/python/framework/sysconfig.py

@@ -21,8 +21,8 @@ import importlib.util
 
 import oneflow
 from oneflow.python.oneflow_export import oneflow_export
-from oneflow.python_gen.sysconfig import generated_compile_flags
 from typing import List
+import oneflow_api
 
 
 @oneflow_export("sysconfig.get_include")
@@ -40,7 +40,12 @@ def get_compile_flags() -> List[str]:
     flags = []
     flags.append("-I{}".format(get_include()))
     flags.append("-DHALF_ENABLE_CPP11_USER_LITERALS=0")
-    flags.extend(generated_compile_flags)
+    if oneflow_api.flags.with_cuda():
+        flags.append("-DWITH_CUDA")
+    if oneflow_api.flags.use_cxx11_abi():
+        flags.append("-D_GLIBCXX_USE_CXX11_ABI=1")
+    else:
+        flags.append("-D_GLIBCXX_USE_CXX11_ABI=0")
     return flags
 
 

oneflow/python/test/ops/test_broadcast_normal.py

@@ -101,8 +101,8 @@ def compare_with_tensorflow_grad(
     input_maxval=10,
     out_rtol=1e-5,
     out_atol=1e-5,
-    diff_rtol=1e-5,
-    diff_atol=1e-5,
+    diff_rtol=1e-4,
+    diff_atol=1e-3,
 ):
     assert device_type in ["gpu", "cpu"]