Add note: SOLVEPNP_EPNP will be used instead of SOLVEPNP_DLS and instead of SOLVEPNP_UPNP.

* Contribute a tutorial on building for NVIDIA Tegra

* Clean-up and fixes based on feedback.

Don't run binary operations for floating-point numbers
norm() will fail with NAN result.

Fixed OpenVX wrapper for Mat::convertTo() (#7794)

* fixed for cases of unrolled (w*h x 1) matrices

* more error handling

Several fixes for OpenVX wrappers (#7782)

* Several fixes added to ivx.hpp:
* `#include <VX/vxu.h>`
* `vx_keypoint_t` support added
* type check fixed
* fixed Node::create() for >6 args in c++98
* fixed Array::copyFrom() if array is empty
* type check changed; Array::addItem methods changed

Add new 5x5 gaussian blur kernel for CV_8UC1 format,
it is 50% ~ 70% faster than current ocl kernel in the perf test.
Signed-off-by: Li Peng <peng.li@intel.com>

Operations inside for loops are unnecessary. They are repeated anyway just below for loops

Core_ArithmMask.uninitialized