Conflicts:
	.gitignore
	doc/tutorials/objdetect/cascade_classifier/cascade_classifier.rst
	modules/gpu/src/match_template.cpp
	modules/imgproc/include/opencv2/imgproc/imgproc.hpp
	modules/ocl/include/opencv2/ocl/ocl.hpp
	modules/ocl/perf/perf_precomp.hpp

Previously we incorrectly assumed data was aligned in 16 bytes, which should be 32.

it conflicts with cv::detail

changed default parameters in CreateERFiltersNM{1,2} functions. They are not needed now in the sample example.

begin_x[1] is not the second component of the element, but the element
after the one pointed to begin_x. When begin_x points to the last
element, that line overwrites data past the end of the allocation, which,
during my tests, happened to contain the reference count for the matrix.
Hilarity ensues.