我有一个CUDA C代码,当我尝试编译它时,nvcc抱怨一个未定义的标识符错误,但它确实退出了变量!
extern "C"
void vRand3f_cuda (vec3f *d_v, int n)
{
curandState *d_states;
CUDA_CALL (cudaMalloc ((void **)&d_states, n * sizeof (curandState)));
dim3 gDim (1);
dim3 bDim (n);
set_seed<<<gDim, bDim>>> (d_states, time (NULL), n);
vRand3f_cuda_generate<<<gDim, bDim>>> (d_v, d_states, n);
CUDA_CALL (cudaFree (d_states));
}
__global__ void set_seed (curandState *states, unsigned long seed, int n)
{
int idx = threadIdx.x + blockIdx.x * gridDim.x;
if (idx >= n)
return ;
curand_init (seed, idx, 0, &states[idx]);
}
__global__ void vRand3f_cuda_generate (vec3f *v, curandState *states, int n)
{
int idx = threadIdx.x + blockIdx.x * gridDim.x;
if (idx >= n)
curandState localS = states[idx];
double s, x, y;
s = 2.;
while (s > 1.)
{
x = 2. - curand_uniform_double (&localS) - 1.;
y = 2. - curand_uniform_double (&localS) - 1.;
s = x * x + y * y;
}
v[idx].z = 1. - 2. * s;
s = 2. * sqrt (1. - s);
v[idx].x = s * x;
v[idx].y = s * y;
states[idx] = localS;
}
我使用以下行编译:
nvcc -m64 -g -G `pkg-config --cflags glib-2.0` -gencode arch=compute_20,code=sm_20 -gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 -I/Developer/NVIDIA/CUDA-5.0/include -I. -o vec3-cuda.o -c vec3-cuda.cu
我得到以下内容:
vec3-cuda.cu(58):警告:变量“localS”已声明,但从未 参考
vec3-cuda.cu(64):错误:标识符“localS”未定义
vec3-cuda.cu(74):错误:标识符“localS”未定义
知道这里发生了什么吗?我在OSX上使用CUDA 5.
谢谢。
答案 0 :(得分:3)
这是错误的:
if (idx >= n)
curandState localS = states[idx];
你可能意味着这样的事情:
if (idx >= n) return;
curandState localS = states[idx];
如上所述,if子句中定义的变量范围是if子句的本地范围。由于您从未在if子句中引用它,因此您将收到第一个警告。在其他地方,当您尝试在if子句的范围之外引用它时会出现错误。