0
我已經將一個項目從CUDA 4.0,Nsight 2.0升級到CUDA 4.2,Nsight 2.2,在CUDA 4.2上重新編譯CUDPP。我可以在升級後正確使用斷點和步驟設備代碼,但是當我的代碼調用cudppScan()函數時,即使函數正確執行操作並返回正確的前綴和值,Nsight也會報告訪問衝突。在沒有Nsight的情況下運行釋放或調試代碼不會產生問題。這個問題可以通過下面的代碼片段重現。對於CUDA 4.2中的Nsight調試實例,Nsight 2.2是否正確調試CUDPP 2.0?
#include <cassert>
#include <cudpp/cudpp.h>
//Cudpp testing
int main(int argc, char* argv[])
{
#define ARRAY_SIZE 32
unsigned int uiArraySize = sizeof(unsigned int)*ARRAY_SIZE;
unsigned int h_auiTest[ARRAY_SIZE];
for(int i = 0 ; i < ARRAY_SIZE ; ++i)
{
h_auiTest[i] = 1;
}
unsigned int *d_auiTest;
GPUMALLOC((void**)&d_auiTest , uiArraySize);
TOGPU(d_auiTest , h_auiTest , uiArraySize);
unsigned int *d_auiOut;
GPUMALLOC((void**)&d_auiOut , uiArraySize);
CUDPPHandle cudppContext;
CUDPPHandle scanPlan;
CUDPP_SAFE_CALL(cudppCreate(&cudppContext));
CUDPPConfiguration scanConfig;
scanConfig.algorithm = CUDPP_SCAN;
scanConfig.datatype = CUDPP_UINT;
scanConfig.op = CUDPP_ADD;
scanConfig.options = CUDPP_OPTION_EXCLUSIVE | CUDPP_OPTION_FORWARD;
cudppPlan(cudppContext , &scanPlan, scanConfig, ARRAY_SIZE , 1, 0);
cudppScan(scanPlan, d_auiOut, d_auiTest , ARRAY_SIZE);
FROMGPU(h_auiTest , d_auiOut , uiArraySize);
for(int i = 0 ; i < ARRAY_SIZE ; ++i)
{
assert(h_auiTest[i] == i);
}
}