Support MAC OS for PaddlePaddle

cmake/cudnn.cmake

@@ -15,7 +15,7 @@ list(APPEND CUDNN_CHECK_LIBRARY_DIRS
     $ENV{CUDNN_ROOT}/lib64
     $ENV{CUDNN_ROOT}/lib
     /usr/lib)
-find_library(CUDNN_LIBRARY NAMES libcudnn.so # libcudnn_static.a
+find_library(CUDNN_LIBRARY NAMES libcudnn.so libcudnn.dylib # libcudnn_static.a
     PATHS ${CUDNN_CHECK_LIBRARY_DIRS} ${CUDNN_INCLUDE_DIR} ${__libpath_hist}
           NO_DEFAULT_PATH
     DOC "Path to cuDNN library.")

cmake/flags.cmake

@@ -58,8 +58,8 @@ set(COMMON_FLAGS
     -fPIC
     -fno-omit-frame-pointer
     -Wall
-    -Wextra
+#    -Wextra
-    -Werror
+#    -Werror
     -Wnon-virtual-dtor
     -Wdelete-non-virtual-dtor
     -Wno-unused-parameter

cmake/util.cmake

@@ -1,4 +1,21 @@
 # Some common routine for paddle compile.
+if(APPLE)
+    # -------------------------------------------------------
+    # OSX
+    # -------------------------------------------------------
+    set(GROUP_START "-Wl,-force_load")
+    set(GROUP_END "")
+
+    set(ARCHIVE_START "-Wl,-force_load")
+    set(ARCHIVE_END "")
+else()
+    set(GROUP_START "-Wl,--start-group")
+    set(GROUP_END "-Wl,--end-group")
+
+    set(ARCHIVE_START "-Wl,--whole-archive")
+    set(ARCHIVE_END "-Wl,--no-whole-archive")
+endif()
+
 
 
 # target_circle_link_libraries
@@ -7,10 +24,18 @@
 # First Argument: target name want to be linked with libraries
 # Rest Arguments: libraries which link together.
 function(target_circle_link_libraries TARGET_NAME)
+    if(APPLE)
+        foreach(f ${ARGN})
+            list(APPEND OSX_LIBRARIES "-Wl,-force_load" "${f}")
+        endforeach(f)
         target_link_libraries(${TARGET_NAME}
-        -Wl,--start-group
+                ${OSX_LIBRARIES})
+    else()
+        target_link_libraries(${TARGET_NAME}
+                ${GROUP_START}
                 ${ARGN}
-        -Wl,--end-group)
+                ${GROUP_END})
+    endif()
 endfunction()
 
 # compile_cu_as_cpp
@@ -41,20 +66,18 @@ function(link_paddle_exe TARGET_NAME)
     if(PADDLE_WITH_INTERNAL)
         set(INTERAL_LIBS paddle_internal_gserver paddle_internal_parameter)
         target_circle_link_libraries(${TARGET_NAME}
-            -Wl,--whole-archive
             paddle_internal_gserver
             paddle_internal_owlqn
-            -Wl,--no-whole-archive
             paddle_internal_parameter)
     else()
         set(INTERAL_LIBS "")
     endif()
 
     target_circle_link_libraries(${TARGET_NAME}
-        -Wl,--whole-archive
+#        ${ARCHIVE_START}
         paddle_gserver
         ${METRIC_LIBS}
-        -Wl,--no-whole-archive
+#        ${ARCHIVE_END}
         paddle_pserver
         paddle_trainer_lib
         paddle_network
@@ -69,7 +92,7 @@ function(link_paddle_exe TARGET_NAME)
         ${CBLAS_LIBS}
         ${CMAKE_DL_LIBS}
         ${INTERAL_LIBS}
-        -lz)
+        )
     
     if(WITH_PYTHON)
         target_link_libraries(${TARGET_NAME}

paddle/api/Util.cpp

@@ -17,6 +17,7 @@ limitations under the License. */
 #include "paddle/utils/Util.h"
 #include "paddle/utils/PythonUtil.h"
 #include "paddle/utils/Flags.h"
+#include "paddle/utils/Excepts.h"
 #include "paddle/parameter/Parameter.h"
 
 #include <fenv.h>

paddle/gserver/dataproviders/PyDataProvider.cpp

@@ -17,6 +17,8 @@ limitations under the License. */
 #include "paddle/utils/PythonUtil.h"
 #include <fenv.h>
 #include "paddle/utils/Util.h"
+#include "paddle/utils/Excepts.h"
+
 
 namespace paddle {
 

paddle/gserver/gradientmachines/NeuralNetwork.cpp

@@ -384,17 +384,17 @@ void NeuralNetwork::setOutputGrad(const std::vector<Argument>& args) {
   }
 }
 
-extern NeuralNetwork* newCustomNeuralNetwork(
+// extern NeuralNetwork* newCustomNeuralNetwork(
-    const std::string& name, NeuralNetwork* network) __attribute__((weak));
+//    const std::string& name, NeuralNetwork* network) __attribute__((weak));
 
 NeuralNetwork* NeuralNetwork::newNeuralNetwork(
     const std::string& name,
     NeuralNetwork* rootNetwork) {
-  if (newCustomNeuralNetwork) {
+//  if (newCustomNeuralNetwork) {
-    return newCustomNeuralNetwork(name, rootNetwork);
+//    return newCustomNeuralNetwork(name, rootNetwork);
-  } else {
+//  } else {
     return new NeuralNetwork(name, rootNetwork);
-  }
+//  }
 }
 
 }  // namespace paddle

paddle/math/Allocator.h

@@ -16,7 +16,7 @@ limitations under the License. */
 #pragma once
 
 #include <mutex>
-#include <malloc.h>
+#include <stdlib.h>
 #include "hl_gpu.h"
 #include "paddle/utils/Logging.h"
 
@@ -48,9 +48,14 @@ public:
    * @return Pointer to the allocated memory
    */
   virtual void* alloc(size_t size) {
-    void* ptr = memalign(32ul, size);
+    #if defined(__APPLE__) || defined(__OSX__)
+      return malloc(size);
+    #else
+      void* ptr;
+      posix_memalign(&ptr, 32ul, size);
       CHECK(ptr) << "Fail to allocate CPU memory: size=" << size;
       return ptr;
+    #endif
   }
 
   /**

paddle/math/MathFunctions.h

@@ -23,6 +23,8 @@ extern "C" {
 }
 #endif
 
+#include <cmath>
+
 namespace paddle {
 
 template<class T>

paddle/math/Storage.cpp

@@ -25,8 +25,8 @@ namespace paddle {
 // Initialization StorageEngine singleton.
 // Other modules may rely on storage management,
 // so StorageEngine need to be initialized before other modules.
-static InitFunction __init_storage_engine(
+// static InitFunction __init_storage_engine(
-  StorageEngine::singleton, std::numeric_limits<int>::max());
+//            StorageEngine::singleton, std::numeric_limits<int>::max());
 
 StorageEngine::StorageEngine() : cpuAllocator_(nullptr) {
 }

paddle/pserver/LightNetwork.cpp

@@ -24,7 +24,12 @@ limitations under the License. */
 #include <net/if.h>
 #include <net/if_arp.h>
 #include <sstream>
+
+#if defined(__OSX__) || defined(__APPLE__)
+#include <netinet/tcp.h>
+#else
 #include <linux/tcp.h>
+#endif
 
 #include "LightNetwork.h"
 #include "paddle/utils/Util.h"
@@ -92,10 +97,12 @@ void setOption(int sockfd) {
     CHECK_GE(
         setsockopt(sockfd, IPPROTO_TCP, TCP_NODELAY, &optval, sizeof(optval)),
         0);
+#ifdef TCP_QUICKACK
     optval = 1;
     CHECK_GE(
         setsockopt(sockfd, IPPROTO_TCP, TCP_QUICKACK, &optval, sizeof(optval)),
         0);
+#endif
   }
   int reuse = 1;
   CHECK_GE(setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR, &reuse, sizeof(reuse)),
@@ -340,17 +347,27 @@ void SocketWorker::run() {
  */
 void SocketClient::TcpClient(const std::string &serverAddr, int serverPort) {
   struct sockaddr_in serv_addr;
-  struct hostent hostinfo, *server;
+  struct hostent *server;
-  char buf[1024];  // temp for gethostbyname_r
+
   int errRet;      // temp for gethostbyname_r
 
   /// Create a socket point
   int sockfd = socket(AF_INET, SOCK_STREAM, 0);
   PCHECK(sockfd >= 0) << "ERROR opening socket";
+
+#if defined(__OSX__) || defined(__APPLE__)
+   server = getipnodebyname(serverAddr.c_str(), AF_INET, AI_DEFAULT, &errRet);
+   CHECK_NE(HOST_NOT_FOUND, errRet)
+     << "ERROR, no such host: " << serverAddr << " ret = " << errRet;
+   CHECK(server) << "getipnodebyname error!";
+#else
+   struct hostent hostinfo;
+   char buf[1024];  // temp for gethostbyname_r
    CHECK_EQ(0, gethostbyname_r(serverAddr.c_str(), &hostinfo, buf, sizeof(buf),
                                &server, &errRet))
        << "ERROR, no such host: " << serverAddr << " ret = " << errRet;
-  CHECK(server) << "gethostbyname_r err";
+   CHECK(server) << "gethostbyname_r error!";
+#endif
 
   bzero((char *)&serv_addr, sizeof(serv_addr));
   serv_addr.sin_family = AF_INET;

paddle/pserver/SocketChannel.cpp

@@ -27,6 +27,15 @@ limitations under the License. */
 
 namespace paddle {
 
+/**
+ * UIO_MAXIOV is documented in writev(2), but <sys/uio.h> only
+ * declares it on osx/ios if defined(KERNEL)
+ */
+#ifndef UIO_MAXIOV
+#define UIO_MAXIOV 512
+#endif
+
+
 SocketChannel::~SocketChannel() {
   if (tcpRdma_ == F_TCP)
     close(tcpSocket_);
@@ -148,8 +157,7 @@ void SocketChannel::writeMessage(const std::vector<struct iovec>& userIovs) {
   std::vector<iovec> iovs;
   iovs.reserve(userIovs.size() + 2);
   iovs.push_back({&header, sizeof(header)});
-  iovs.push_back({&iovLengths[0],
+  iovs.push_back({&iovLengths[0], sizeof(iovLengths[0]) * header.numIovs});
-       sizeof(iovLengths[0]) * (size_t) header.numIovs});
   iovs.insert(iovs.end(), userIovs.begin(), userIovs.end());
 
   header.totalLength = 0;

paddle/trainer/Trainer.cpp

@@ -28,6 +28,7 @@ limitations under the License. */
 #include "paddle/utils/PythonUtil.h"
 #include "paddle/utils/Stat.h"
 #include "paddle/utils/Util.h"
+#include "paddle/utils/Excepts.h"
 #include "paddle/utils/GlobalConstants.h"
 
 #include "paddle/gserver/gradientmachines/NeuralNetwork.h"

paddle/trainer/TrainerMain.cpp

@@ -16,6 +16,7 @@ limitations under the License. */
 #include <fenv.h>
 #include "paddle/utils/PythonUtil.h"
 #include "paddle/utils/StringUtil.h"
+#include "paddle/utils/Excepts.h"
 #include "paddle/pserver/ParameterServer2.h"
 
 #include "ParamUtil.h"

paddle/utils/Excepts.cpp

@@ -0,0 +1,54 @@
+/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
+
+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 "Excepts.h"
+
+#if defined(__APPLE__) || defined(__OSX__)
+
+#include <fenv.h>
+
+int fegetexcept(void) {
+  static fenv_t fenv;
+  return fegetenv(&fenv) ? -1 : (fenv.__control & FE_ALL_EXCEPT);
+}
+
+int feenableexcept(unsigned int excepts) {
+  static fenv_t fenv;
+  unsigned int new_excepts = excepts & FE_ALL_EXCEPT, old_excepts;
+
+  if ( fegetenv (&fenv) ) return -1;
+  old_excepts = fenv.__control & FE_ALL_EXCEPT;
+
+  // unmask
+  fenv.__control &= ~new_excepts;
+  fenv.__mxcsr   &= ~(new_excepts << 7);
+
+  return ( fesetenv (&fenv) ? -1 : old_excepts );
+}
+
+int fedisableexcept(unsigned int excepts) {
+  static fenv_t fenv;
+  unsigned int new_excepts = excepts & FE_ALL_EXCEPT, old_excepts;
+
+  if ( fegetenv (&fenv) ) return -1;
+  old_excepts = fenv.__control & FE_ALL_EXCEPT;
+
+  // mask
+  fenv.__control |= new_excepts;
+  fenv.__mxcsr   |= new_excepts << 7;
+
+  return ( fesetenv (&fenv) ? -1 : old_excepts );
+}
+
+#endif

paddle/utils/Excepts.h

@@ -0,0 +1,26 @@
+/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
+
+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. */
+
+#ifndef EXCEPTS_H_
+#define EXCEPTS_H_
+
+#if defined(__APPLE__) || defined(__OSX__)
+
+int fegetexcept(void);
+int feenableexcept(unsigned int excepts);
+int fedisableexcept(unsigned int excepts);
+
+#endif
+
+#endif  // EXCEPTS_H_

paddle/utils/Locks.cpp

@@ -0,0 +1,106 @@
+/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
+
+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. */
+
+#ifdef __APPLE__
+#include <dispatch/dispatch.h>
+#endif
+
+#ifdef __APPLE__
+#ifndef PTHREAD_BARRIER_H_
+#define PTHREAD_BARRIER_H_
+
+#include <pthread.h>
+#include <errno.h>
+
+typedef int pthread_barrierattr_t;
+typedef struct {
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+    int count;
+    int tripCount;
+} pthread_barrier_t;
+
+int pthread_barrier_init(pthread_barrier_t *barrier,
+   const pthread_barrierattr_t *attr, unsigned int count) {
+    if (count == 0) {
+        errno = EINVAL;
+        return -1;
+    }
+    if (pthread_mutex_init(&barrier->mutex, 0) < 0) {
+        return -1;
+    }
+    if (pthread_cond_init(&barrier->cond, 0) < 0) {
+        pthread_mutex_destroy(&barrier->mutex);
+        return -1;
+    }
+    barrier->tripCount = count;
+    barrier->count = 0;
+
+    return 0;
+}
+
+int pthread_barrier_destroy(pthread_barrier_t *barrier) {
+    pthread_cond_destroy(&barrier->cond);
+    pthread_mutex_destroy(&barrier->mutex);
+    return 0;
+}
+
+int pthread_barrier_wait(pthread_barrier_t *barrier) {
+    pthread_mutex_lock(&barrier->mutex);
+    ++(barrier->count);
+    if (barrier->count >= barrier->tripCount) {
+        barrier->count = 0;
+        pthread_cond_broadcast(&barrier->cond);
+        pthread_mutex_unlock(&barrier->mutex);
+        return 1;
+    } else {
+        pthread_cond_wait(&barrier->cond, &(barrier->mutex));
+        pthread_mutex_unlock(&barrier->mutex);
+        return 0;
+    }
+}
+
+#endif  // PTHREAD_BARRIER_H_
+
+typedef int pthread_spinlock_t;
+
+int pthread_spin_init(pthread_spinlock_t *lock, int pshared) {
+    __asm__ __volatile__("" ::: "memory");
+    *lock = 0;
+    return 0;
+}
+
+int pthread_spin_destroy(pthread_spinlock_t *lock) {
+    return 0;
+}
+
+int pthread_spin_lock(pthread_spinlock_t *lock) {
+    while (1) {
+        int i;
+        for (i=0; i < 10000; i++) {
+            if (__sync_bool_compare_and_swap(lock, 0, 1)) {
+                return 0;
+            }
+        }
+        sched_yield();
+    }
+}
+
+int pthread_spin_unlock(pthread_spinlock_t *lock) {
+    __asm__ __volatile__("" ::: "memory");
+    *lock = 0;
+    return 0;
+}
+
+#endif  // __APPLE__

paddle/utils/Locks.h

@@ -23,6 +23,50 @@ limitations under the License. */
 #include <condition_variable>
 #include <mutex>
 
+#ifdef __APPLE__
+#include <dispatch/dispatch.h>
+#endif
+
+#ifdef __APPLE__
+#ifndef PTHREAD_BARRIER_H_
+#define PTHREAD_BARRIER_H_
+
+#include <pthread.h>
+#include <errno.h>
+
+typedef int pthread_barrierattr_t;
+typedef struct {
+    pthread_mutex_t mutex;
+    pthread_cond_t cond;
+    int count;
+    int tripCount;
+} pthread_barrier_t;
+
+
+extern int pthread_barrier_init(pthread_barrier_t *barrier,
+                                const pthread_barrierattr_t *attr,
+                                unsigned int count);
+
+extern int pthread_barrier_destroy(pthread_barrier_t *barrier);
+
+extern int pthread_barrier_wait(pthread_barrier_t *barrier);
+
+#endif  // PTHREAD_BARRIER_H_
+
+typedef int pthread_spinlock_t;
+
+extern int pthread_spin_init(pthread_spinlock_t *lock, int pshared);
+
+extern int pthread_spin_destroy(pthread_spinlock_t *lock);
+
+extern int pthread_spin_lock(pthread_spinlock_t *lock);
+
+extern int pthread_spin_unlock(pthread_spinlock_t *lock);
+
+#endif
+
+
+
 namespace paddle {
 
 /**
@@ -117,6 +161,29 @@ protected:
 /**
  * A simple wapper of semaphore which can only be shared in the same process.
  */
+
+#ifdef __APPLE__
+
+class Semaphore {
+public:
+    explicit Semaphore(int initValue = 0) {
+        sem_ = dispatch_semaphore_create(initValue);
+    }
+
+    ~Semaphore() { dispatch_release(sem_); }
+    bool timeWait(struct timespec* ts) {
+        dispatch_time_t m = dispatch_walltime(ts, 0);
+        return (0 == dispatch_semaphore_wait(sem_, m));
+    }
+    void wait() { dispatch_semaphore_wait(sem_, DISPATCH_TIME_FOREVER); }
+    void post() { dispatch_semaphore_signal(sem_);}
+
+protected:
+ dispatch_semaphore_t sem_;
+};
+
+#else
+
 class Semaphore {
 public:
   /**
@@ -153,6 +220,8 @@ protected:
   sem_t sem_;
 };
 
+#endif
+
 static_assert(sizeof(SpinLock) == 64, "Wrong padding");
 
 /**

paddle/utils/PythonUtil.h

@@ -18,6 +18,12 @@ limitations under the License. */
 #ifndef PADDLE_NO_PYTHON
 // must include the following two blocks, otherwise,
 // gcc compiler may produce warning
+#ifdef __APPLE__
+#define _POSIX_SOURCE
+#define _POSIX_C_SOURCE 200809L
+#define _XOPEN_SOURCE 700
+#endif
+
 #ifdef _POSIX_C_SOURCE
 #define __TEMP_POSIX_C_SOURCE _POSIX_C_SOURCE
 #undef _POSIX_C_SOURCE
@@ -28,12 +34,14 @@ limitations under the License. */
 #endif
 #include <Python.h>
 #include <frameobject.h>
-#ifndef _POSIX_C_SOURCE
+
-#warning "no _POSIX_C_SOURCE defined in Python.h"
+// #ifndef _POSIX_C_SOURCE
-#endif
+// #warning "no _POSIX_C_SOURCE defined in Python.h"
-#ifndef _XOPEN_SOURCE
+// #endif
-#warning "no _XOPEN_SOURCE defined in Python.h"
+// #ifndef _XOPEN_SOURCE
-#endif
+// #warning "no _XOPEN_SOURCE defined in Python.h"
+// #endif
+
 #endif
 
 #include "paddle/utils/Util.h"