Fix compiling errors at CPU && OSX support

Dragon/CMakeLists.txt

@@ -22,11 +22,11 @@ option(WITH_CUDA_FP16              "Set ON to use FP16"  ON)
 # Set your 3rdparty
 set(3RDPARTY_DIR  ${PROJECT_SOURCE_DIR}/../3rdparty)
 
-# set your python "interpreter" if necessary
+# Set your python "interpreter" if necessary
 # if not, a default interpreter will be used
 # here, provide several examples:
-# set(PYTHON_EXECUTABLE /usr/bin/python) # Linux, OS
-# set(PYTHON_EXECUTABLE /X/anaconda/bin/python) # Linux, Anaconda
+# set(PYTHON_EXECUTABLE /usr/bin/python) # Linux & OSX, Builtin Python
+# set(PYTHON_EXECUTABLE /X/anaconda/bin/python) # Linux & OSX, Anaconda
 # set(PYTHON_EXECUTABLE X:/Anaconda/python) # Win, Anaconda
 
 # Set CUDA compiling architecture
@@ -37,9 +37,10 @@ set(CUDA_ARCH    -gencode arch=compute_30,code=sm_30
                  -gencode arch=compute_60,code=sm_60
                  -gencode arch=compute_70,code=sm_70)
 
-# Set CUDNN Library Dir if necessary (Linux Only)
+# Set CUDNN Library Dir if necessary (Linux/OSX Only)
 # For Win, Recommend to use ``3RDPARTY_DIR/lib``
-set(CUDNN_LIBRARY_DIR /usr/local/cuda/lib64)
+set(CUDNN_LIBRARY_DIR /usr/local/cuda/lib64) # Linux
+# set(CUDNN_LIBRARY_DIR /usr/local/cuda/lib) # OSX
 
 # ---------------- User Config ----------------
 
@@ -178,7 +179,7 @@ endif()
 if(UNIX)
     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -s -fPIC")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -s -w -fPIC -O3 -m64 -std=c++11")
-    if (WITH_OMP)
+    if (WITH_OMP AND (NOT APPLE))
         set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fopenmp")
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp")
     endif()

Dragon/include/core/common.h

@@ -24,6 +24,7 @@
 #include <unordered_set>
 #include <algorithm>
 #include <mutex>
+#include <functional>
 
 #include "core/types.h"
 #include "protos/dragon.pb.h"

Dragon/include/core/tensor.h

@@ -12,8 +12,6 @@
 #ifndef DRAGON_CORE_TENSOR_H_
 #define DRAONG_CORE_TENSOR_H_
 
-#include <vector>
-
 #include "core/common.h"
 #include "core/mixedmem.h"
 

Dragon/include/utils/cuda_device.h

@@ -13,20 +13,22 @@
 #define DRAGON_UTILS_CUDA_DEVICE_H_
 
 #ifdef WITH_CUDA
-
 #include <cuda_runtime.h>
 #include <cublas.h>
 #include <curand.h>
 #include <cuda.h>
+#endif
 
 #ifdef WITH_MPI_NCCL
 #include <nccl.h>
-#endif  // WITH_MPI_NCCL
+#endif
 
 #include "core/common.h"
 
 namespace dragon {
 
+#ifdef WITH_CUDA
+
 static const int CUDA_NUM_THREADS = 1024;
 //  We do have a server with 10 GPUs :-)
 #define MAX_GPUS 10

Dragon/modules/cxx/CMakeLists.txt

@@ -3,7 +3,7 @@ message(STATUS "Found CXX Module: ${CMAKE_CURRENT_LIST_DIR}")
 FILE(GLOB_RECURSE MODULE_FILES *.h *.hpp *.c *.cpp *.cu *.cc)
 FILE(GLOB_RECURSE SRC_FILES ../../src/*.c ../../src/*.cpp ../../src/*.cu ../../src/*.cc)
 
-# ---[ complier
+# ---[ Target
 if (WITH_CUDA) 
     CUDA_ADD_LIBRARY(${PROJECT_NAME}_cxx SHARED ${MODULE_FILES} ${SRC_FILES})
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_cxx  ${CUDA_LIBRARIES} ${CUDA_cublas_LIBRARY} ${CUDA_curand_LIBRARY})
@@ -11,14 +11,13 @@ else ()
     ADD_LIBRARY(${PROJECT_NAME}_cxx SHARED ${MODULE_FILES} ${SRC_FILES})
 endif()
 
-
-# ---[ link basics
+# ---[ Linker(Basics)
 FILE(GLOB targets ${3RDPARTY_LIBS}/*.so ${3RDPARTY_LIBS}/*.lib) 
 foreach(target ${targets})
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_cxx  ${target})
 endforeach()
 
-# ---[ link optional libs
+# ---[ Linker(OptionalLibs)
 if (UNIX AND WITH_CUDNN)
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_cxx cudnn)
 endif()
@@ -29,7 +28,7 @@ if (UNIX AND WITH_MPI_NCCL)
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_cxx nccl)
 endif()
 
-# ---[ link platforms
+# ---[ Linker(Platforms)
 if(UNIX)
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_cxx protobuf pthread)
 endif()
@@ -37,7 +36,7 @@ if(WIN32)
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_cxx shlwapi.lib)
 endif()
 
-set_target_properties(${PROJECT_NAME}_cxx PROPERTIES OUTPUT_NAME dragon_cxx)
+SET_TARGET_PROPERTIES(${PROJECT_NAME}_cxx PROPERTIES OUTPUT_NAME dragon_cxx)
 
-# ---[ install
-install (TARGETS ${PROJECT_NAME}_cxx DESTINATION ${PROJECT_BINARY_DIR}/../lib)
\ No newline at end of file
+# ---[ Install
+INSTALL(TARGETS ${PROJECT_NAME}_cxx DESTINATION ${PROJECT_BINARY_DIR}/../lib)
\ No newline at end of file

Dragon/modules/python/CMakeLists.txt

@@ -7,7 +7,7 @@ endif()
 FILE(GLOB_RECURSE MODULE_FILES *.h *.hpp *.c *.cpp *.cu *.cc)
 FILE(GLOB_RECURSE SRC_FILES ../../src/*.c ../../src/*.cpp ../../src/*.cu ../../src/*.cc)
 
-# ---[ complier
+# ---[ Target
 if (WITH_CUDA) 
     CUDA_ADD_LIBRARY(${PROJECT_NAME}_python SHARED ${MODULE_FILES} ${SRC_FILES})
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_python  ${CUDA_LIBRARIES} ${CUDA_cublas_LIBRARY} ${CUDA_curand_LIBRARY})
@@ -15,14 +15,13 @@ else ()
     ADD_LIBRARY(${PROJECT_NAME}_python SHARED ${MODULE_FILES} ${SRC_FILES})
 endif()
 
-
-# ---[ link basics
+# ---[ Linker(Basics)
 FILE(GLOB targets ${3RDPARTY_LIBS}/*.so ${3RDPARTY_LIBS}/*.lib) 
 foreach(target ${targets})
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_python  ${target})
 endforeach()
 
-# ---[ link optional libs
+# ---[ Linker(OptionalLibs)
 if (UNIX AND WITH_CUDNN)
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_python cudnn)
 endif()
@@ -33,15 +32,19 @@ if (UNIX AND WITH_MPI_NCCL)
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_python nccl)
 endif()
 
-# ---[ link platforms
+# ---[ Linker(Platforms)
 if(UNIX)
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_python protobuf pthread)
 endif()
 if(WIN32)
     TARGET_LINK_LIBRARIES(${PROJECT_NAME}_python shlwapi.lib)
+    TARGET_LINK_LIBRARIES(${PROJECT_NAME}_python ${PYTHON_LIBRARIES})
+endif()
+if (APPLE)
+  SET_TARGET_PROPERTIES(${PROJECT_NAME}_python PROPERTIES LINK_FLAGS "-undefined dynamic_lookup")
 endif()
 
-set_target_properties(${PROJECT_NAME}_python PROPERTIES OUTPUT_NAME dragon)
+SET_TARGET_PROPERTIES(${PROJECT_NAME}_python PROPERTIES OUTPUT_NAME dragon)
 
-# ---[ install
-install (TARGETS ${PROJECT_NAME}_python DESTINATION ${PROJECT_BINARY_DIR}/../lib)
\ No newline at end of file
+# ---[ Install
+INSTALL(TARGETS ${PROJECT_NAME}_python DESTINATION ${PROJECT_BINARY_DIR}/../lib)
\ No newline at end of file

Dragon/modules/python/py_cuda.h

@@ -14,17 +14,15 @@
 
 #include "dragon.h"
 
-#ifdef WITH_CUDA
-
 inline PyObject* IsCUDADriverSufficientCC(PyObject* self, PyObject* args) {
+#ifdef WITH_CUDA
     int count;
     cudaError_t err = cudaGetDeviceCount(&count);
     if (err == cudaErrorInsufficientDriver) return PyBool_FromLong(0);
     return PyBool_FromLong(1);
-}
-
 #else
     CUDA_NOT_COMPILED;
 #endif
+}
 
 #endif    // DRAGON_PYTHON_PY_MPI_H_
\ No newline at end of file

Dragon/python/setup.py

@@ -14,13 +14,17 @@ def find_packages(root_dir):
             if filename == '__init__.py':
                 packages.append(root_dir)
 
+
 def find_modules():
     dragon_c_lib_win32 = '../lib/dragon.dll'
-    dragon_c_lib_other = '../lib/libdragon.so'
+    dragon_c_lib_linux = '../lib/libdragon.so'
+    dragon_c_lib_darwin = '../lib/libdragon.dylib'
     if os.path.exists(dragon_c_lib_win32):
         shutil.copy(dragon_c_lib_win32, 'dragon/libdragon.pyd')
-    elif os.path.exists(dragon_c_lib_other):
-        shutil.copy(dragon_c_lib_other, 'dragon/libdragon.so')
+    elif os.path.exists(dragon_c_lib_linux):
+        shutil.copy(dragon_c_lib_linux, 'dragon/libdragon.so')
+    elif os.path.exists(dragon_c_lib_darwin):
+        shutil.copy(dragon_c_lib_darwin, 'dragon/libdragon.so')
     else:
         print('ERROR: Unable to find modules. built Dragon using CMake.')
         sys.exit()
@@ -32,11 +36,13 @@ def find_resources():
     others = []
     return c_lib + protos + others
 
+
 find_packages('dragon')
 find_modules()
 
+
 setup(name = 'dragon',
-      version='0.2.2.2',
+      version='0.2.2.3',
       description = 'Dragon: A Computation Graph Virtual Machine Based Deep Learning Framework',
       url='https://github.com/seetaresearch/Dragon',
       author='Ting Pan',

Dragon/src/core/mixedmem.cc

@@ -164,7 +164,9 @@ const Map<string, string> MixedMemory::info() const {
     }
     s2s["mem_at"] = _state_;
     if (cpu_ptr_) s2s["CPU"] = "0";
+#ifdef WITH_CUDA
     if (cuda_ptr_) s2s["CUDA"] = dragon_cast<string, int>(CUDA_POINTER_DEVICE(cuda_ptr_));
+#endif
     return s2s;
 }