模板函数调用其他函数 [英] Template function calls to other functions

问题描述

我理解模板函数通常要在头文件中声明和定义。

I understand the template functions usually are to be declared and defined in header files.

我的是模板函数调用其他函数。

The problem I am having is that my template function makes calls to other functions. The prototypes of those other functions are in the same header file before the template function itself.

这部分代码

//header.h

template <int ignoreAdetection>
__global__ void MCMLKernel(SimState d_state, GPUThreadStates tstates)
{
  // photon structure stored in registers
  PhotonStructGPU photon;

  // random number seeds
  UINT64 rnd_x;
  UINT32 rnd_a;

  // Flag to indicate if this thread is active
  UINT32 is_active;

  // Restore the thread state from global memory.
  RestoreThreadState(&d_state, &tstates, &photon, &rnd_x, &rnd_a, &is_active);
...
...
}

RestoreThreadState 是从此模板函数调用的几个函数中的第一个。其他的在一个for循环内调用。

The function RestoreThreadState is the first of several functions called from this template function. The others are called within a for loop.

我不知道这个模板函数应该或不应该在头文件。

I am not sure if this template function should or should not be in the header file. If should be in a header file, how do I call those other functions?

在MCMLKernel的实例化过程中，我从编译器得到的错误：

The errors I am getting from the compiler during the instantiation of MCMLKernel:

• 错误：缺少显式类型（假设为int）


• 错误：变量RestoreThreadState


• 错误：类型SimState *的值不能用于初始化类型为int的实体


• 错误：expected a


• 警告：声明与以前的RestoreThreadState不兼容

• error: explicit type is missing ("int" assumed)
• error: variable "RestoreThreadState" may not be initialized
• error: a value of type "SimState *" cannot be used to initialize an entity of type "int"
• error: expected a ")"
• warning: declaration is incompatible with previous "RestoreThreadState"

所有这些功能都是CUDA内核函数。 MCMLKernel 是一个 __ global __ 内核，其调用的其余函数是 __ device __ 内核。我使用Nsight Eclipse版和计算能力1.3 GPU（四个特斯拉C1060卡）。

Additional details. All this functions are CUDA kernel functions. MCMLKernel is a __global__ kernel and the rest of the functions it calls are __device__ kernels. I am using Nsight Eclipse Edition and compute capability 1.3 GPUs (four Tesla C1060 cards).

推荐答案

@Eugene提供了答案问题。

@Eugene provided the answer to this question.

我创建了调用函数的原型，如下所示：

I created the prototype of the called-to function like this

__device__ void RestoreThreadState(SimState *d_state, GPUThreadStates *tstates,
                               PhotonStructGPU *photon,
                               UINT64 *rnd_x, UINT32 *rnd_a,
                               UINT32 *is_active);

但是，用Eugene给出的例子应该是这样的（不是PROTOTYPE！ ）

But, with an example given by Eugene it should be like this (NOT A PROTOTYPE! See other answer)

__device__ void RestoreThreadState(SimState *d_state, GPUThreadStates *tstates,
                               PhotonStructGPU *photon,
                               UINT64 *rnd_x, UINT32 *rnd_a,
                               UINT32 *is_active){}

在代码中我调用函数实际返回像这样

Next in the code I call to functions which actually return values like this

__device__ int HitBoundary(PhotonStructGPU *photon);

从相同的 MCMLKernel 模板函数调用。它给我一个警告：

Called from the same MCMLKernel template function. It is giving me a warning:

• 警告：在非空函数HitBoundary结束时缺少返回语句

UPDATE：在另一个源文件 kernel.cu 中，我有以下声明和定义（为什么我有多个定义问题）

UPDATE: In another source file kernel.cu I have the following declaration and definition (reason why I have multiple definition problems):

__device__ void RestoreThreadState(SimState *d_state, GPUThreadStates *tstates,
                                   PhotonStructGPU *photon,
                                   UINT64 *rnd_x, UINT32 *rnd_a,
                                   UINT32 *is_active)
{
  UINT32 tid = blockIdx.x * NUM_THREADS_PER_BLOCK + threadIdx.x;

  *rnd_x = d_state->x[tid];
  *rnd_a = d_state->a[tid];

  photon->x = tstates->photon_x[tid];
  photon->y = tstates->photon_y[tid];
  photon->z = tstates->photon_z[tid];
  photon->ux = tstates->photon_ux[tid];
  photon->uy = tstates->photon_uy[tid];
  photon->uz = tstates->photon_uz[tid];
  photon->w = tstates->photon_w[tid];
  photon->sleft = tstates->photon_sleft[tid];
  photon->layer = tstates->photon_layer[tid];

  *is_active = tstates->is_active[tid];
}

IN总结：我有四个源文件

IN SUMMARY: I have four source files

1. main.cu


2. kernel.cu b $ b
3. rng.cu


4. mem.cu

1. main.cu
2. kernel.cu
3. rng.cu
4. mem.cu

除 main.cu 之外的每个源文件都有一个相关的标题

Each source file, except main.cu has an associated header

1. kernel.cuh
   
2. rng.cuh / strong>

1. kernel.cuh
2. rng.cuh
3. mem.cuh

其中，我要转发声明要在 main.cu 中使用的函数。

where I want to forward-declare functions to use in main.cu.

一切正常，直到我进入模板函数，该函数调用 kernel.cu rng.cu 。

Everything is fine until I get to the template function which calls functions from kernel.cu and rng.cu.

这篇关于模板函数调用其他函数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！