Question

我正在开发GPU / OpenCL NBody代码。我使用AMD APP SDK的OpenGL渲染粒子位置。运行代码时，我有随机分段错误。

总而言之，我有一个GLWidget，我在其中进行OpenGL渲染。生成初始位置后，我将在此GLWidget中渲染它们。之后，我运行模拟，并在每一步计算下一个位置并在GLwidget中显示它们。我的问题是，有时候，如果我在模拟运行时点击参数GUI的“生成初始条件”按钮，我就会出现分段错误：

这是回溯：

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff4a46cd7 in memcpy () from /lib/libc.so.6
(gdb) bt
#0  0x00007ffff4a46cd7 in memcpy () from /lib/libc.so.6
#1  0x00007fffeda2da64 in ?? () from /usr/lib/x86_64-linux-gnu/dri/fglrx_dri.so
#2  0x00007fffedbba74a in ?? () from /usr/lib/x86_64-linux-gnu/dri/fglrx_dri.so
#3  0x00007fffedbba9af in ?? () from /usr/lib/x86_64-linux-gnu/dri/fglrx_dri.so
#4  0x00007fffed9c56e4 in ?? () from /usr/lib/x86_64-linux-gnu/dri/fglrx_dri.so
#5  0x00007fffed17371d in ?? () from /usr/lib/x86_64-linux-gnu/dri/fglrx_dri.so
#6  0x000000000040b185 in GLWidget::createVBO() ()
#7  0x000000000040b3c9 in GLWidget::draw() ()
#8  0x000000000040c36d in GLWidget::processCurrent() ()
...

这是createVBO例程：

void GLWidget::createVBO()
{   
  GLuint vbo;
  int memSize = sizeof(cl_double4) * 4 * Galaxy->getNumParticles();
  glGenBuffers(1, &vbo);
  glBindBuffer(GL_ARRAY_BUFFER, vbo);
  glBufferData(GL_ARRAY_BUFFER, memSize, Galaxy->pos, GL_DYNAMIC_DRAW);
}

段错误发生在glBufferData(GL_ARRAY_BUFFER, memSize, Galaxy->pos, GL_DYNAMIC_DRAW);

我不明白为什么会这样。当我按下“生成IC”按钮时，我删除已分配的Galaxy->pos数组并创建一个新数组。

以下是我在“生成IC”例程中所做的事情：

  //Clean Galaxy already existing 
  if (parent->widget_2->isGalaxyExist)
  { 
    if (parent->widget_2->animation)
      parent->resetSimu();
    parent->widget_2->Galaxy->cleanup();
  }

使用cleanup例程（我删除pos数组）：

int NBody::cleanup()
{
  if (glEvent)
    clReleaseEvent(glEvent);

  // Releases OpenCL resources (Context, Memory etc.)
  cl_int status;

  if (hasRunKernel)
  {
  status = clFinish(commandQueue);
  CHECK_OPENCL_ERROR(status, "clFinish failed.(commandQueue)");

  status = clReleaseKernel(kernel);
  CHECK_OPENCL_ERROR(status, "clReleaseKernel failed.(kernel)");

  status = clReleaseProgram(program);
  CHECK_OPENCL_ERROR(status, "clReleaseProgram failed.(program)");

  status = clReleaseMemObject(currPos);
  CHECK_OPENCL_ERROR(status, "clReleaseMemObject failed.(currPos)");

  status = clReleaseMemObject(currVel);
  CHECK_OPENCL_ERROR(status, "clReleaseMemObject failed.(currVel)");

  status = clReleaseMemObject(newPos);
  CHECK_OPENCL_ERROR(status, "clReleaseMemObject failed.(newPos)");

  status = clReleaseMemObject(newVel);
  CHECK_OPENCL_ERROR(status, "clReleaseMemObject failed.(newVel)");

  status = clReleaseCommandQueue(commandQueue);
  CHECK_OPENCL_ERROR(status, "clReleaseCommandQueue failed.(commandQueue)");

  status = clReleaseContext(context);
  CHECK_OPENCL_ERROR(status, "clReleaseContext failed.(context)");

  hasRunKernel = false;
  }

  // Release program resources 
  delete [] pos;
  delete [] vel;
  delete [] initPos;
  delete [] initVel;
  delete [] devices;
  // Delete current instance
  delete this;

  return NBODY_SUCCESS;
}

乍一看，你能看出什么是错的，或者给我一个关于这个段错误的线索。最令人讨厌的是，错误是随机发生的，而不是每次都执行。

Answer 1

这个计算是否正确？

int memSize = sizeof(cl_double4) * 4 * Galaxy->getNumParticles();

特别是“* 4”：sizeof（cl_double4）已经考虑了向量的四个元素。

Answer 2

这样的崩溃表明通过glBufferData OpenGL API函数调用的驱动程序代码中的越界访问。检查传递给glBufferData的参数是否正确，即给glBufferData读取的长度是否作为数据参数传递的内存范围内。

GPU / OpenCL / OpenGL代码上的随机分段错误

2 个答案: