使用cuda的OpenCV 3.0.0：要求启动的资源太多

Question

我从最新源（2014.09.22）编译了OpenCV 3.0.0

我试图使用opencv_contrib/modules/xfeatures2d/samples/surf_matcher.cpp但没有成功。

我的代码：

int surf() {
   cv::cuda::printShortCudaDeviceInfo( cv::cuda::getDevice( ) );

   cv::cuda::SURF_CUDA surf;
   std::cerr << "- 1 -" << std::endl;
   cv::cuda::GpuMat img1, keypoints1GPU, descriptors1GPU;
   std::cerr << "- 2 -" << std::endl;
   cv::Mat im1( imread( "shapes.jpg", cv::IMREAD_GRAYSCALE ));
   std::cerr << "- 3 - load(): " << ( im1.data ? "OK" : "failed" ) << std::endl;
   img1.upload( im1 );
   std::cerr << "- 4 -" << std::endl;
   surf( img1, cv::cuda::GpuMat(), keypoints1GPU, descriptors1GPU );
   std::cerr << "- 5 -" << std::endl;

控制台：

Device 0:  "GeForce 8600 GT"  256Mb, sm_11 (not Fermi), 32 cores,
   Driver/Runtime ver.6.50/6.50
- 1 -
- 2 -
- 3 - load(): OK
- 4 -
OpenCV Error: Gpu API call (too many resources requested for launch) in
   cv::cuda::device::surf::compute_descriptors_gpu,
   file D:/opencv_contrib/modules/xfeatures2d/src/cuda/surf.cu, line 947
D:/opencv_contrib/modules/xfeatures2d/src/cuda/surf.cu:947: error: (-217)
   too many resources requested for launch in function
   cv::cuda::device::surf::compute_descriptors_gpu

仅供参考，这里是surf.cu的代码：

void compute_descriptors_gpu(
   PtrStepSz<float4>  descriptors,
   const float *      featureX,
   const float *      featureY,
   const float *      featureSize,
   const float *      featureDir,
   int                nFeatures )
{
   if( descriptors.cols == 64 ) {
      ...            
   }
   else {
      compute_descriptors_128<<<nFeatures, dim3(32, 16)>>>(
         descriptors, featureX, featureY, featureSize, featureDir );
      cudaSafeCall( cudaGetLastError()); <<<<<<<<<<<<<<<<<<<<< line 947

操作系统：WINDOWS 7 64位

VisualStudio Express 2013，目标是Win32，调试模式