chevron error

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi!!
I don’t know why this code is error…T_T
->VectorAdd<<<65535,512>>>(dev_A,dev_B,dev_R,size);
second chevron is error…
Please give me some solution!! please!!!

#include <stdio.h>
#include <cutil_inline.h>
#include <cuda_runtime_api.h>
#include <device_launch_parameters.h>

global void VectorAdd(int *a, int *b,int size){

int tid = blockIdx.x * blockDim.x+ threadIdx.x;
a[tid] += b[tid];

}
int main(){
const int size = 51265535;
const int BufferSize = sizesizeof(int);

int *InputA;
int *InputB;
int *Result;

InputA = (int*)malloc(BufferSize);
InputB = (int*)malloc(BufferSize);
Result = (int*)malloc(BufferSize);

int i=0;

for(int i=0; i<size; i++){
	InputA[i]=i;
	InputB[i]=i;
	Result[i]=0;
}

int* dev_A;
int* dev_B;
int* dev_R;

cudaMalloc((void**)&dev_A,size*sizeof(int));
cudaMalloc((void**)&dev_B,size*sizeof(int));
cudaMalloc((void**)&dev_R,size*sizeof(int));

cudaMemcpy(dev_A,InputA,size*sizeof(int),cudaMemcpyHostToDevice);
cudaMemcpy(dev_B,InputB,size*sizeof(int),cudaMemcpyHostToDevice);

VectorAdd<<<65535,512>>>(dev_A,dev_B,dev_R,size);

cudaMemcpy(Result,dev_R,size*sizeof(int),cudaMemcpyDeviceToHost);

for(int i=0; i<5; i++){
	printf("Result[%d] : %d \n",i,Result[i]);
}
printf("........\n");
for(i=size-5; i<size; i++){
	printf("Result[%d] : %d \n",i,Result[i]);
}

cudaFree(dev_A);
cudaFree(dev_B);
cudaFree(dev_R);

free(InputA);
free(InputB);
free(Result);

return 0;

}

2

You call your kernel with four parameters (dev_A,dev_B,dev_R,size), but in your kernel code, you only ask for three paramters (int *a, int *b,int size). Furthermore, your result a[tid] maps to dev_A, but you memcopy the content of dev_R to Result.

3

Thank you for recommand my problem.
I fixed code again,you told me
but…it was still wrong…

global void VectorAdd(int *a, int *b,int *c,int size){

int tid = blockIdx.x * blockDim.x+ threadIdx.x;
c[tid]=a[tid] + b[tid];

}

I correct the code that but it is still wrong…
Could you give me the solution again?

4

If I understand your original post correctly, the compiler is flagging a syntax error on the first kernel invocation? That would indicate you are trying to compile this code as regular C or C++ code using your host compiler (which does not understand triple angular brackets, since that is part of CUDA but not part of C/C++). If you place your CUDA code in a file with a .cu file extension, and compile that file with nvcc, this particular problem should go away.