Igor Podgainoi authored May 14, 2024 and Michael Platings committed May 14, 2024

On some platforms such as macOS, the compiler can fail with error:
non-constant-expression cannot be narrowed from type [...]

Reword the description of the KLEIDICV_LIMIT_SVE2_TO_SELECTED_ALGORITHMS
option.

The performance uplift from these operations is inconsistent across
devices so remove them from the HAL until a consistent uplift can be
achieved.

-Werror causes an error if an ignorable warning is found. This is
beneficial during development because it ensures that warnings are
fixed, but it is harmful for deployment since it can turn warnings that
may be spurious into blockers for a user to build the project.

Therefore don't specify -Werror in CMakeLists that must be used to build
the project. Instead specify it in outer build scripts that a user will
typically bypass.

The idiomatic way to specify -Werror in CMake 3.24 and later is to use
CMAKE_COMPILE_WARNING_AS_ERROR so use that.

Versions were decided according to those provided by Ubuntu 20.04

This change adds a workaround for Clang 10 & 11 which had incomplete
SVE2 support.

On GCC 9.4 fix this error:
error: parameter ‘callback’ set but not used [-Werror=unused-but-set-parameter]

OpenCV 4.9 has a bug that cv_hal_minMaxIdx is called regardless of the
number of channels in the image even though the interface has no way to
pass the channel count to it.

A fix is planned in OpenCV 4.10 but as yet it is unstable. Therefore
disable the implementation in KleidiCV for now.

With some toolchains the lack of the cstdint header caused compilation
failure.

If the library is built with OpenCV the undefined macros cause an error.

Move the documentation to a common "doc" folder to make docs more
discoverable.

Split documentation into multiple pages.

Separate Android & Linux build documentation. Android is the primary
platform so the build docs now prioritises instructions for Android.

Add patch as build prerequisite.

Move build.sh documentation inside the script itself. Otherwise remove
it from the documentation as it has been found to create more confusion
than it solves except for advanced users.

Mark KLEIDICV_CHECK_BANNED_FUNCTIONS as internal since there's no
particular need for users to touch it.

Since the warnings were added the project's quality bar has been raised
greatly and the warnings are no longer appropriate.

Previously KLEIDICV_ENABLE_SVE2 and KLEIDICV_ENABLE_SVE2_SELECTIVELY
were mutually exclusive. Now KLEIDICV_ENABLE_SVE2 enables or disables
SVE2 code paths overall, and KLEIDICV_LIMIT_SVE2_TO_SELECTED_ALGORITHMS
is a refinement of that option.

The intent of this change is merely to make the options more ergonomic.
It's now only necessary to set
KLEIDICV_LIMIT_SVE2_TO_SELECTED_ALGORITHMS=OFF in order to enable all
SVE2 code paths, instead of setting both
KLEIDICV_ENABLE_SVE2_SELECTIVELY=OFF and KLEIDICV_ENABLE_SVE2=ON.

Fix a typo and remove the BUILD_ANDROID_PROJECTS CMake variable which
is not used in OpenCV.

cv_hal_convertTo is proposed to be added to OpenCV but for now is not
present. To avoid a compilation error when building with the head of the
4.x branch, don't build kleidicv_convertTo_with_fallback if
cv_hal_convertTo is not defined.

The conversions were previously not implemented for Neon. The
functionality and the tests have now been removed entirely. It is
intended that conversions will be re-added, except with Neon
implementations included.

The lack of Neon implementation causes complexity for integrating the
library. The functionality will be re-added in future, including a Neon
implementation.

The intent of the cast was likely to ensure numeric values were printed
instead of printing as char. Unary + achieves this better via integer
promotion.

Michael Platings authored Apr 18, 2024 and Denes Tarjan committed Apr 19, 2024

* CMake's standard <PROJECT-NAME>_SOURCE_DIR can be used instead of
  creating a new OPENCV_SOURCE_DIR. See
  https://cmake.org/cmake/help/latest/variable/PROJECT-NAME_SOURCE_DIR.html
* Connect to the new HAL function that has been added to OpenCV that
  matches the Gaussian blur provided by KleidiCV.
  See https://github.com/opencv/opencv/pull/25397
* Rename cv_hal_transpose to cv_hal_transpose2d to match
  https://github.com/opencv/opencv/pull/25342

* Add a readme
* Base the dev Docker image on the image used in CI.
* Upgrade from Bash to Zsh, with oh-my-zsh.
* Move ccache directory outside the repo, to keep the repo more clean.
* Use the real username of the current user.
* Use real path names inside the container.
* Allow using vim instead of nano.
* Enable Git clone with SSH.
* Install some nice-to-have command line utilities.
* Install a couple of nice-to-have VSCode extensions.