getting started problems

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I previously installed nsight eclipse on OS X 10.8, with Cuda 5.5, and the samples worked fine. I had to reinstall the whole system, and now is trying again, but I am having problems.

Nsight Eclipse Edition

Version: 5.0.0
CUDA Driver Version: 5.5.25
GPU Driver Version: 8.12.47 310.40.00.05f01

First: Advanced Quick Sort Sample is giving me linking and compilation errors:

In the project build properties, I choose to generate PTX code 3.5 and GPU Code 3.5, then I had this linking problem:

ptxas fatal : Unresolved extern function ‘cudaStreamCreateWithFlags’

search forums have suggested adding -dlink, I did that in the NVCC Linker libraries, but didn’t show up in the linking command.

Then I realised that I can’t generate GPU code 3.5 because no compatible device on the system, and only 1.2 was fine in the properties dialog, but generated the following errors:

…/src/cdpAdvancedQuicksort.cu(145): error: identifier “__ballot” is undefined

…/src/cdpAdvancedQuicksort.cu(165): error: identifier “__shfl” is undefined

…/src/cdpAdvancedQuicksort.cu(223): error: calling a global function(“big_bitonicsort”) from a global function(“qsort_warp”) is only allowed on the compute_35 architecture or above

…/src/cdpAdvancedQuicksort.cu(234): error: calling a global function(“qsort_warp”) from a global function(“qsort_warp”) is only allowed on the compute_35 architecture or above

…/src/cdpAdvancedQuicksort.cu(244): error: calling a global function(“bitonicsort”) from a global function(“qsort_warp”) is only allowed on the compute_35 architecture or above

…/src/cdpAdvancedQuicksort.cu(260): error: calling a global function(“big_bitonicsort”) from a global function(“qsort_warp”) is only allowed on the compute_35 architecture or above

…/src/cdpAdvancedQuicksort.cu(270): error: calling a global function(“qsort_warp”) from a global function(“qsort_warp”) is only allowed on the compute_35 architecture or above

…/src/cdpAdvancedQuicksort.cu(277): error: calling a global function(“bitonicsort”) from a global function(“qsort_warp”) is only allowed on the compute_35 architecture or above

Does this mean that my GPU device will not be able to run this sample? Is there a workaround?

Second, I though this should be like Eclipse environment, where SDK documentation help is integrated in the editor, and code continuation and list of properties will show up. Is there a way to integrate CUDA libraries documentation to nsight?

I appreciate your help very much,

2

Shuffle operations (__shfl) are available only for devices with compute capability 3.x.

The vote intrinsic __ballot is available only for devices with compute capability 2.x or higher.

Calling a global function from a global function is known as dynamic parallelism and is only allowed on the compute_35 architecture or above, as the error message says.

There is no workaround: your GPU won’t be able to run that example.