Use 4x FMA chains to sum on SIMD 128 FP64 targets. On
x86 this showed about 1.4x improvement.

For PPC, do a full multiply (32x32->64b), convert to DP
then accumulate. This may be slightly less precise for
some inputs. But is 1.5x faster than the above which
is about 1.5x than the FMA above for ~2.5x speedup.

* Unroll multiply and add instructions in dotProd_32f - 35% faster.

* Eliminate unnecessary v_reduce_sum instructions.

* Improve support for Windows Embedded Compact

* Remove redundant set(WINCE true) and format CMake

* Added support for the ONNX "ReduceMean" Layer. (as this is the same as the GlobalAveragePool)

* Add ReduceMean test

* Fix ONNX importer

* Fix ReduceMean

* Add assert

* Split test

* Fix split test

  * use SOLVE_EPNP for the initial guess

WUI based implementation to initUndistortRectifyMap (#14994)

* Add initUndistortRectifyMap performance test

* Move cv namespace boundaries

* Add wide universal intrinsics based implementation to initUndistortRectifyMap

* Dispatch undistort

* Fix dumpToFile

* Add test

* Fix test

* Reduce store gathering pressures - speeds thresholds by up to 20%

* Rename temporary histogram array and initialize so that MACOSX builder is happy

- writeRaw(): support structs
- readRaw(): 'len' is buffer limit in bytes (documentation is fixed)

* Support Deconvolution3D on IE backend

* Add test tag

* Fix tests

may cause crashes on ARM platform

* Support Conv3D on OCV backend

* Add header

* Add perf tests

* Support pool3d

* Enable Resnet34_kinetics on OCV backend

* Add test

* Fix conv

* Optimize Conv2D