Atomic operation Getting atomicAdd support

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I am using CUDA1.0 and i want to do some atomic operation in a memory location at global area. My display card is 8800GTS. I changed the custom build setup to

$(CUDA_BIN_PATH)\nvcc.exe -arch sm_11 -ccbin “$(VCInstallDir)bin” -c -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/O2,/Zi,/MT -I"$(CUDA_INC_PATH)" -I./ -I…/…/common/inc -o $(ConfigurationName)\template.obj template.cu

But the program is giving some different output than what i expect.

kernel
global void
testKernel( int* g_odata)
{
// Block index
int bx = blockIdx.x;
int by = blockIdx.y;

// Thread index
int tx = threadIdx.x;
int ty = threadIdx.y;
if(tx==0&&bx==0)
{
  g_odata[0]=0;
}
int nBlocksize = 16;
int nStart = bx * ceil((float)65536/nBlocksize)  + tx * ceil((float)((65536/nBlocksize)/nBlocksize));
for( int i = nStart; i <= nStart+ceil((float)((65536/nBlocksize)/nBlocksize)); i=i+1 )
{
   g_odata[0] = 1.0f; 
   //__syncthreads();
}

}

host
void
runTest( int argc, char** argv)
{
CUT_DEVICE_INIT();
int* pCpuOutData = (int*)malloc( 256256sizeof(float));
int* pOutData;
CUDA_SAFE_CALL( cudaMalloc( (void**) &pOutData, 256 * 256 * sizeof(int)));
dim3 grid(16,1);
dim3 thread(16,1);
testKernel<<<grid,thread>>>(pOutData);

CUDA_SAFE_CALL( cudaMemcpy( pCpuOutData, pOutData, 256 * 256 * sizeof(int),
                cudaMemcpyDeviceToHost) );    



printf("%d",pCpuOutData[0]);
CUDA_SAFE_CALL( cudaFree(pOutData));   
free( pCpuOutData );

}

The output is some junk value like 11731320…

Please help me.

2

Alas, the 8800GTS doesn’t support atomic operations, you need a card with compute model 1.1 for that, like the 8600 (G86) or 8800GT (G92)

3

is there any table which specifies the card number like 1.1 or 1.0? if so where?

4

The CUDA FAQ includes a list of supported GPUs and their compute version:
[url=“http://forums.nvidia.com/index.php?showtopic=36286”]http://forums.nvidia.com/index.php?showtopic=36286[/url]

We will be updating this for CUDA 1.1 shortly.