【发布时间】：2011-02-20 11:10:58
【问题描述】：
我正在做一个关于光子映射的项目。我编写了光线追踪器部分，它在 CPU 上成功运行。现在我在 GPU 上做同样的事情（通过 ssh）。
我收到以下错误
nvcc -c -lSDL -lGL -lGLU AntTweakBar.a gpuRayTracer.cu gpuRayTracer.cu(44): error: identifier "raytracer" is undefined gpuRayTracer.cu(53): error: no suitable conversion function from“Float3”到“void *”存在
gpuRayTracer.cu(55): error: no suitable conversion function from“Float3”到“void *”存在
gpuRayTracer.cu(76): error: identifier "GPUsub" is undefined gpuRayTracer.cu(77): error: identifier "GPUnormalize" is undefined gpuRayTracer.cu(78): error: identifier "GPUcross" is undefined gpuRayTracer.cu(80): error: calling a host function from a设备/_全局_函数不允许
gpuRayTracer.cu(90): error: identifier "GPUmul" is undefined gpuRayTracer.cu(95): error: calling a host function from a设备/_全局_函数不允许
gpuRayTracer.cu(95): error: identifier "GPUadd" is undefined gpuRayTracer.cu(192): error: calling a host function from a设备/_全局_函数不允许
15 errors detected in the compilation of"/tmp/tmpxft_0000432c_00000000-4_gpuRayTracer.cpp1.ii" . make: * [gpuRayTracer.o] 错误 2
gpuRayTracer.cu 第 44,53、55 行（错误）标记在下面的代码中
下面使用的Float3是一个包含3个浮点变量（x,y,z坐标）的结构
void Scene::GPUrayTracer(){
Object *d_objectList[OBJ_MAX];
GLubyte * d_pixels;
int *d_Width, *d_Height;
Float3 *d_eye,*d_lookAt;
int *d_objectCount;
size_t size1=sizeof(Float3);
size_t size2=sizeof(int);
size_t size3=sizeof(GLubyte);
//size_t size4=sizeof(Object);
cudaMalloc(&d_eye,size1);
cudaMalloc(&d_lookAt,size1);
cudaMemcpy(d_eye,&this->eye,size1,cudaMemcpyHostToDevice);
cudaMemcpy(d_lookAt,&this->lookAt,size1,cudaMemcpyHostToDevice);
cudaMalloc(&d_objectCount,size2);
cudaMemcpy(d_objectCount,&this->objectCount,size2,cudaMemcpyHostToDevice);
cudaMalloc(&d_Width,size2);
cudaMalloc(&d_Height,size2);
cudaMemcpy(d_Width,&this->screenWidth,size2,cudaMemcpyHostToDevice);
cudaMemcpy(d_Height,&this->screenHeight,size2,cudaMemcpyHostToDevice);
cudaMalloc(&d_pixels,size3);
cudaMemcpy(d_pixels,&this->pixels,size3,cudaMemcpyHostToDevice);
cudaMalloc((void **)&d_objectList,
(sizeof(this->objectList)));
cudaMemcpy(d_objectList,
&this->objectList,
sizeof(this->objectList),cudaMemcpyHostToDevice);
line 44:raytracer<<<1,500>>>(d_pixels,d_Width,d_Height,
d_objectList,d_eye,d_lookAt);
cudaMemcpy((this->objectList),&d_objectList,sizeof(this-
>objectList),cudaMemcpyDeviceToHost);
cudaMemcpy(this->pixels,&d_pixels,size3,cudaMemcpyDeviceToHost);
cudaMemcpy((int *)this->screenWidth,&d_Width,size2,cudaMemcpyDeviceToHost);
cudaMemcpy((int *)this->screenHeight,&d_Height,size2,cudaMemcpyDeviceToHost);
cudaMemcpy((int *)this->objectCount,&d_objectCount,size2,cudaMemcpyDeviceToHost);
cudaMemcpy(
line:53 (void *)this->eye,
(void *)&d_eye,sizeof(d_eye),cudaMemcpyDeviceToHost);
line:55 cudaMemcpy(this->lookAt,(void *)&d_lookAt,sizeof(d_lookAt),cudaMemcpyDeviceToHost);
}
__global__ void raytracer( unsigned char *out_data,const int screenWidth,const int screenHeight,Object * objectList,Float3 eye,Float3 lookAt,int objectCount)
{
int x = blockDim.x * BLOCK_SIZE + threadIdx.x;
int y = blockDim.y * BLOCK_SIZE + threadIdx.y;
[b]//code goes here[/b]
}
__device__ float GPUffminf(float a, float b){
if(a<b)
return a;
return b;
}
__device__ float GPUffmaxf(float a, float b){
if(a>b)
return a;
return b;
}
__device__ float GPUmag(Float3 a){
float res;
res=a.x*a.x+a.y*a.y+a.z*a.z;
res=sqrt(res);
return res;
}
__device__ Float3 GPUnormalize(Float3 a){
Float3 res;
float magn=mag(a);
if(magn!=0){
magn=(float)1.0/magn;
res.x=a.x*magn;
res.y=a.y*magn;
res.z=a.z*magn;
return res;
}
return a;
}
__device__ Float3 GPUcross(Float3 a ,Float3 b){
Float3 res;
res.x=a.y*b.z-a.z*b.y;
res.y=a.z*b.x-a.x*b.z;
res.z=a.x*b.y-a.y*b.x;
return res;
}
__device__ float GPUdot(Float3 a,Float3 b){
return (float)(a.x*b.x + a.y*b.y + a.z*b.z);
}
__device__ Float3 GPUsub(Float3 a,Float3 b){
Float3 res;
res.x=a.x-b.x;
res.y=a.y-b.y;
res.z=a.z-b.z;
return res;
}
__device__ Float3 GPUadd(Float3 a,Float3 b){
Float3 res;
res.x=a.x+b.x;
res.y=a.y+b.y;
res.z=a.z+b.z;
return res;
}
__device__ Float3 GPUmul(Float3 a,float b){
Float3 res;
res.x=a.x*b;
res.y=a.y*b;
res.z=a.z*b;
return res;
}
代码有问题吗？？
除此之外我还有几个问题
*.cu/.cpp文件的编译顺序..有关系吗？？ *是否应该只从 main.cpp 调用内核？？ *如果是这样，.cu 文件应该只包含全局/设备函数吗？
【问题讨论】：
-
首先，首先在文件顶部定义所有设备功能的原型，如下所示：“__global__ void raytracer(unsigned char,int,int,Object*,Float3,Float3,int) ;"