如何在使用Boost的lockfree spsc_queue时进行编译？ [英] How do I compile while using Boost's lockfree spsc_queue?

问题描述

我试图在CUDA中编写一个程序，它将在主机端使用Boost的单个生产者/单个消费者队列，但是一旦我包括& boost，我遇到编译器错误/lockfree/spsc_queue.hpp> 。

此时，我在设备上什么也不做， $ c> spsc_queue 对象。

最初我使用Boost 1.54，使用GCC 4.8.4编译，并使用所有的标志它们作为CUDA 7.0 makefile的一部分以及以下自定义标志： -Xcompiler -fopenmp -lgomp -std = c ++ 11 -lpthread 。这些在变量MYFLAGS中，如下所示：

  main.o：main.cu 
 $（EXEC）$ （NVCC）$（INCLUDES）$（ALL_CCFLAGS）$（GENCODE_FLAGS）-c $（MYFLAGS）main.cu。 
  

然后在连接步骤中，我有

  $（EXEC）$（NVCC）$（ALL_LDFLAGS）$（GENCODE_FLAGS）-O2 $（MYFLAGS）-o runnable $（OBJECTS）$（LIBRARIES）。 
  

包含标题时，我收到此错误：

  /usr/include/boost/utility/detail/result_of_iterate.hpp:135:75：错误：
无效使用qualified-name std :: allocator_traits< _Alloc> :: propagate_on_container_swap'
  

我在Boost中搜索可能的错误， href =http://stackoverflow.com/questions/30779322/compile-error-when-using-boostlockfreespsc-queue-is-it-a-bug-in-boost>这一个。

我升级到Boost 1.58，现在我收到不同的错误：

  boost_1_58_0 / boost / lockfree / spsc_queue.hpp（352）：错误：函数调用中的参数太少。 
  

spsc_queue.hpp 中的问题代码部分的复位例程，但是我还没有声明一个对象。

  void reset（void） 
 {
 if（！boost :: has_trivial_destructor< T> :: value）{
 //确保调用所有析构函数！ 
 
 T dummy_element; 
 while（pop（dummy_element））
 {} 
} else {
 write_index_.store（0，memory_order_relaxed）; 
 read_index_.store（0，memory_order_release）; 
} 
} 
  

因此，我的问题： Boost的lockfree spsc_queue？

解决方案

问题是 nvcc 不像例如那样成熟 g ++ 当涉及到复杂的C ++代码，如Boost。一些Boost库使用 nvcc ，其他编译失败; boost :: lockfree 似乎属于后面的组。

一个可能的解决方案是拆分主机和设备代码，并编译所有包含头 nvcc 的主机代码不能理解 g ++ 。

I'm trying to write a program in CUDA that will use Boost's single-producer/single-consumer queue on the host side, but I run into compiler errors as soon as I include <boost/lockfree/spsc_queue.hpp>.

At this point, I am doing nothing on the device, nor am I creating an spsc_queue object. The errors occurred as soon as I tried to compile with the above header.

Originally I was using Boost 1.54, compiling with GCC 4.8.4, and using all of the flags that come as part of the CUDA 7.0 makefile along with the following custom flags: -Xcompiler -fopenmp -lgomp -std=c++11 -lpthread. These are in the variable MYFLAGS, seen below:

main.o: main.cu
    $(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -c $(MYFLAGS) main.cu.

Then in the linking step, I have

$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -O2 $(MYFLAGS) -o runnable $(OBJECTS) $(LIBRARIES).

As soon as I included the header, I received this error:

/usr/include/boost/utility/detail/result_of_iterate.hpp:135:75: error:
    invalid use of qualified-name std::allocator_traits<_Alloc>::propagate_on_container_swap’

I searched for possible bugs in Boost, and I discovered this one. Unfortunately, that's not at all the issue I'm having.

I upgraded to Boost 1.58, and now I receive a different error:

boost_1_58_0/boost/lockfree/spsc_queue.hpp(352): error: too few arguments in function call.

The problem code in spsc_queue.hpp is part of a "reset" routine, but again, I haven't even declared an object yet.

void reset(void)
{
    if ( !boost::has_trivial_destructor<T>::value ) {
        // make sure to call all destructors!

        T dummy_element;
        while (pop(dummy_element))
        {}
    } else {
        write_index_.store(0, memory_order_relaxed);
        read_index_.store(0, memory_order_release);
    }
}

Hence, my question: How do I compile while using Boost's lockfree spsc_queue? What am I doing wrong?

解决方案

The problem is that nvcc is not as mature as e.g. g++ when it comes to complex C++ code such as Boost. Some Boost libraries work with nvcc, others fail to compile; boost::lockfree seems to belong to latter group.

A possible solution is to split host and device code, and compile all host code which includes headers nvcc does not understand with g++.

这篇关于如何在使用Boost的lockfree spsc_queue时进行编译？的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！