c++ - How to properly link cuda header file with device functions? -

i'm trying decouple code bit , fails. compilation error:

error: calling __host__ function("decoupledcallgpu") __global__ function("kernel") not allowed 

code excerpt:

main.c (has call cuda host function):

#include "cuda_compuations.h" ... computesomething(&var1,&var2); ... 

cuda_computations.cu (has kernel, host master functions , includes header has device unctions):

#include "cuda_computations.h" #include "decoupled_functions.cuh" ... __global__ void kernel(){ ... decoupledcallgpu(&var_kernel); }  void computesomething(int *var1, int *var2){ //allocate memory , etc.. ... kernel<<<20,512>>>(); //cleanup ... } 

decoupled_functions.cuh:

#ifndef _decoupledfunctions_h_ #define _decoupledfunctions_h_  void decoupledcallgpu(int *var);  #endif 

decoupled_functions.cu:

#include "decoupled_functions.cuh"  __device__ void decoupledcallgpu(int *var){   *var=0; }  #endif 

compilation:

nvcc -g --ptxas-options=-v -arch=sm_30 -c cuda_computations.cu -o cuda_computations.o -lcudart

question: why decoupledcallgpu called host function , not kernel supposed to?

p.s.: can share actual code behind if need me to.

add __device__ decorator prototype in decoupled_functions.cuh. should take care of error message seeing.

then you'll need use separate compilation , linking amongst modules. instead of compiling -c compile -dc. , link command need modified. basic example here.

your question bit confusing:

question: why decoupledcallgpu called host function , not kernel supposed to?

i can't tell if you're tripping on english or if there misunderstanding here. actual error message states:

error: calling __host__ function("decoupledcallgpu") __global__ function("kernel") not allowed

this arising due fact within compilation unit (ie. within module, within file being compiled, ie. cuda_computations.cu), description of function decoupledcallgpu() provided in prototype in header:

void decoupledcallgpu(int *var); 

this prototype indicates undecorated function in cuda c, , such functions equivalent to __host__ (only) decorated functions:

__host__ void decoupledcallgpu(int *var); 

that compilation unit has no knowledge of in decoupled_functions.cu.

therefore, when have kernel code this:

__global__ void kernel(){       //<- __global__ function ... decoupledcallgpu(&var_kernel);  //<- appears __host__ function compiler } 

the compiler thinks trying call __host__ function __global__ function, illegal.