Code based on nVidia's CUDA 4.0 SDK did not work with CUDA 5.0

cutil.h and cutil_inline_runtime.h are missing

Others have reported SDK CUDA 5 problems too.

Work around

• In compilation Makefile replace SDK include path with samples include path.
  Eg replace:

  SDK_INC_PATH          = $(CUDA_PATH)/SDK/CUDALibraries/common/inc
  

  with

  SDK_INC_PATH          = $(CUDA_PATH)/samples/common/inc
  

• Replace SDK compilation include files with helper* include files.
  Eg. replace

  #include <cutil.h>
  #include <cutil_inline_runtime.h>
  

  with

  #include <helper_cuda.h>
  #include <helper_timer.h>
  

• Replace cutil* and cut* calls and macros with SDK calls.
  Egs:
  • CUDA_SAFE_CALL with checkCudaErrors

    CUDA_SAFE_CALL(cudaMallocPitch((void**)&g_disparityLeft,&g_floatDispPitch,w*sizeof(float),h));  
    

    checkCudaErrors(cudaMallocPitch((void**)&g_disparityLeft,&g_floatDispPitch,w*sizeof(float),h)); 
    

  • cutilSafeCall with checkCudaErrors
  • cut*Timer with sdk*Timer

    unsigned int timer;
    CUT_SAFE_CALL(cutStartTimer(timer));  
                       .
                       .
                       .
    CUT_SAFE_CALL(cutStopTimer(timer));
    float retval = cutGetTimerValue(timer);
    

    Could be rewritten:

    StopWatchInterface *timer = NULL;
    sdkCreateTimer(&timer);
    sdkStartTimer(&timer);
                       .
                       .
                       .
    sdkStopTimer(&timer);
    float retval = sdkGetTimerValue(&timer);
    

Remaining Problems