NDK r18 drops GCC toolchains and some STL types. We need to choose a
clang toolchain by default when no gcc toolchains are available. Switch
the STL type default to `c++_static` when the old `gnustl_static`
default is not available.
Update the test suite to not run tests for STL types that do not exist.
Also do not expect the gcc toolchain `cpp` tool to be available because
r18 does not provide it. Also teach it to tolerate `gcc -dumpmachine`
output like `arm--linux-android` that differs from the toolchain prefix.
Fixes: #18301
Use `curand_static` to identify directories containing device libraries
because `cublas_device` is deprecated and will be removed in the future.
Issue: #18290
-- Update how the latest OS is determined; scan the location GHS_OS_ROOT and sort it
No longer use registry settings looking for installations
The registry values are assigned in installation order for Green Hills tools not version order
Filter out files from the list of directories (i.e if int1234.zip and int1234 are both in root folder)
Just like the existing WindowsPhone and WindowsStore platform modules
for MSVC, just include the corresponding Windows platform modules.
MinGW tools (both with GCC and Clang) can be used for building for
Windows Store, even though it isn't a very common or simple setup.
Moves `CMAKE_${lang}_COMPILER_PREDEFINES_COMMAND` from linux specific
[Modules/Platform/Linux-GNU.cmake](Modules/Platform/Linux-GNU.cmake) to
[Modules/Compiler/GNU.cmake](Modules/Compiler/GNU.cmake).
This enables compiler predefines generation (in AUTOMOC) on all
platforms that run gcc (and clang).
The `<CMAKE_CUDA_LINK_FLAGS>` placeholder in CUDA rule variables comes
from the `<CMAKE_CXX_LINK_FLAGS>` placeholder from which the CUDA rule
variables were originally derived. It is not a public interface for
adding link flags so no projects should be using it. It is needed for
platform information modules to specify platform-specific link flags for
the language. None of our platform modules set it, so it is unused.
Furthermore, it is broken as currently implemented. Some of the
contexts in which it is used need `-Xlinker` and some do not.
Therefore it is not possible to use the placeholder at all.
Simply remove it for now. If some need for platform-specific CUDA link
flags arises a new solution will be needed.
4597396784 Apple: Use include_guard() within compiler config
e95b3fd9bb Apple: Use CMAKE_EFFECTIVE_SYSTEM_NAME to share compiler info
84f9f63fcc Modules: Introduce CMAKE_EFFECTIVE_SYSTEM_NAME to lookup compiler info
9fa0f2eb56 CMakeFindPackageMode: Perform platform-specific initialization
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !1964
Apple platforms macOS, iOS, tvOS, and watchOS can all share the same
compiler information. Rename `Darwin-*` modules to `Apple-*` and load
them all through `CMAKE_EFFECTIVE_SYSTEM_NAME`. This saves duplication
of 4 * 21 compiler information modules.
Issue: #17870
The `armeabi` ABI is no longer available, so we can no longer use it by
default unconditionally. Instead detect all available ABIs and choose
the oldest arm ABI that is available.
Also update the test suite to account for the lack of `armeabi` support
and pass as of Android NDK r17-beta2.
In some environments, libraries containing device code are installed in
directories searched by the host linker but not passed by nvcc to
nvlink. Make these libraries available by explicitly passing the
host linker search directories during device linking.
Issue: #16317
Revise the logic from commit v3.7.0-rc1~173^2~2 (GNU: Do not use
-fvisibility on AIX or HP-UX, 2016-09-03) to add a version check.
The GCC 7 release notes [1] state that visibility support has been
added for AIX 7.1 and above.
[1] https://gcc.gnu.org/gcc-7/changes.html
Shared libraries linked with CUDA now properly setup booth the correct
install_name but also specify headerpad_max_install_names so that the names
can be changed post build.
When using a real MSVC compiler for `C` or `CXX`, use the version of
that compiler for `MSVC_VERSION`. This is preferred over the MSVC
version that a non-MSVC compiler "simulates".
Fixes: #17468
Since commit v2.8.10~148^2~2 (Modernize MSVC compiler information files,
2012-08-23), the Windows-NMcl platform has been broken.
It's most likely not needed anymore.
Revert commit v3.7.0-rc1~25^2 (VS: Recognize VS/LLVM toolset names as
Clang, 2016-09-28). Since at least LLVM 5.0 the VS integration of the
LLVM toolchain now mimics cl and accepts MSVC-style command-line
arguments (unlike Microsoft Clang/C2).
Fixes: #17193, #17235
Since NDK commit 90ec78ffd96b87cd75d82575587ead14d6494df1 (Remove Clang
toolchain path from setup.mk, 2017-05-31) the `setup.mk` files for Clang
no longer hold the path to the `toolchains/llvm` directory. It has been
the same since NDK r11, so use that as the default.
Fixes: #17253
The VS environment script we load changes the working directory to
`%USERPROFILE%/source` if it exists. Set `VSCMD_START_DIR` to tell
the script to retain the current working directory.
The `crlf` attribute is deprecated in Git. This also changes the given
files to be in the index using LF newlines, but they will be checked
out with CRLF newlines due to the attribute.
