Inlining and customizing sampling functions to reduce memory traffic and compute
Improve calcOrientation implementation.
Using more efficient rounding routines.
Removing unnecessary use of local memory

Looks like we don't actually use them; when creating precompiled headers
with Visual C++, a dummy .cpp is created at build time.

This should allow many algorithms to take advantage of more parallelization
technologies.

Affected functions surf.ocl, pyrlk.ocl and hog.ocl are updated with the change.

Backport from master.

1. Added more sync to reduction.
2. Turned off Image2D feature. Probably its support is not detected correctly.
3. Temporary disabled descriptor tests - can't localize a problem of the ocl descriptor.

changed default parameters of SURF, which improved its performance. Restored bi-linear interpolation in SURF descriptor extractor. Added test for SURF homography + check for non-zero (positive) responses.

modified SIFT to 1) double image before finding keypoints, 2) use floating-point internally instead of 16-bit integers, 3) set the keypoint response to the abs(interpolated_DoG_value). step 1) increases the number of detected keypoints significantly and together with 2) and 3) it improves some detection benchmarks. On the other hand, the stability of the small keypoints is lower, so the rotation and scale invariance tests now struggle a bit. In 2.5 need to make this feature optional and add some more intelligence to the algorithm.

added test that finds a planar object using SIFT.