The `native` architecture compiles for the host's GPUs, but our CI jobs
may may run on hosts with GPUs newer than supported by their version of
the CUDA toolkit. Add an undocumented environment variable to tell
CMake to clamp the native architecture to that supported by the toolkit.
Without this, we may try to compile for architectures not supported by
the CUDA Toolkit, which fails. Since commit d1b48bfabd (CUDA: Add
support for CUDA_ARCHITECTURES=native, 2022-03-04), our CUDA 9.2 CI job
fails when it runs on a CI host with a GPU architecture newer than that
CUDA 9.2 supports. Clamping the architecture level fixes that.
Do not document this clamp behavior publicly, at least for now.
Users can be responsible for building with a CUDA toolkit recent
enough to support their host's GPUs.
Issue: #22375
8abd714176 Help: Clarify that ENVIRONMENT test properties take ;-separated lists
02cf404ace Help: Add advice for dealing with semicolons in lists
c4117d9116 ExternalProject: Document that LIST_SEPARATOR works for CMAKE_ARGS too
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Ben Boeckel <ben.boeckel@kitware.com>
Merge-request: !7066
8abd714176 Help: Clarify that ENVIRONMENT test properties take ;-separated lists
02cf404ace Help: Add advice for dealing with semicolons in lists
c4117d9116 ExternalProject: Document that LIST_SEPARATOR works for CMAKE_ARGS too
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Ben Boeckel <ben.boeckel@kitware.com>
Merge-request: !7066
Since commit 29ea94e17c (BinUtils: Avoid llvm-ar on Apple platforms,
2022-03-03, v3.21.6~1^2) we do not consider `llvm-ar` at all on Apple
platforms. However, there are existing cross-compiling use cases in
which the toolchain has `llvm-ar` but not `ar`. Prior to the
re-ordering in commit cf82300a63 (BinUtils: Clarify search logic and
make it more consistent, 2021-05-27, v3.21.0-rc1~119^2~2), we preferred
`ar` and then `llvm-ar`. Restore the original order for Apple.
Fixes: #23320
Adds additional library names `zlibstat[d]` and `zlibvc[d]` for Windows,
when zlib is built using one of the Visual Studio solutions under
`zlib/contrib/vstudio`.
Adds a `ZLIB_USE_STATIC_LIBS` option that is equivilent to similar
settings in other modules such as FindProtobuf, FindOpenSSL, etc.
Implements #18029 and #23140
Since `CMAKE_ARGS` is used to construct the default `CONFIGURE_COMMAND`
for CMake-based external projects, the `LIST_SEPARATOR` option works for
it too.
CUDA 11.6 added the `nvcc -arch=native` flag to automatically compile
for the host GPUs' architectures. Add support for specifying this
special `native` value in `CMAKE_CUDA_ARCHITECTURES` and
`CUDA_ARCHITECTURES`. During the compiler ABI detection step,
detect the native architectures so we can pass them explicitly
when using Clang or older versions of nvcc.
Fixes: #22375
They are stored in a slightly different place with oneAPI than they
used to be in PSXE.
A similar change was made for Windows by commit 956160bb9a (IRSL: Fix
search for Windows redist files with Intel Classic compiler, 2021-09-23,
v3.22.0-rc1~88^2), which left a comment about the locations relative to
the Classic and oneAPI compilers.
Fixes: #23310
Since commit cf82300a63 (BinUtils: Clarify search logic and make it more
consistent, 2021-05-27, v3.21.0-rc1~119^2~2) we correctly prefer the
more-specific name `llvm-mt` over `mt` when using Clang. However, the
`llvm-mt` tool does not yet support all the flags we need in the
implementation of `vs_link_{exe,dll}`. Prefer plain `mt` for now.
Fixes: #23305
The change in commit cc4da8d13a (IAR/CXX: Fix compatibility with CMP0057
OLD, 2022-01-29, v3.23.0-rc1~46^2) broke the detection of C++ version
because the `IN_LIST` operator cannot work directly on a list but
requires a variable.
Fix logic added by commit 7fdd5128b1 (FindMatlab: Fix version selection
if a version is given, 2021-07-02, v3.22.0-rc1~66^2). Ensure that
`_list_index` is always initialized to -1, akin to `list(FIND)` not
finding a match.
Issue: #22377
9fb1dff070 LINK_LIBRARY: Add features for library support on Apple
93a153bc7f Genx-LINK_LIBRARY: simplify framework features definitions
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !7029
Since commit cf82300a63 (BinUtils: Clarify search logic and make it more
consistent, 2021-05-27, v3.21.0-rc1~119^2~2) we correctly prefer the
more-specific name `llvm-ar` over `ar` when using Clang. However, on
Apple platforms, `llvm-ar` does not generate a symbol table that the
Apple linker accepts. Fall back to `ar` on Apple platforms.
Fixes: #23269
dae3ad08fa Tests: Add cases for CMAKE_CUDA_ARCHITECTURES={all,all-major}
5c1f5357b0 VS: Fix CUDA compiler id with CMAKE_CUDA_ARCHITECTURES={all,all-major}
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !7035
Skip the architecture verification check for these values on Visual
Studio. It cannot be implemented correctly until future work delays the
check to the main compiler test step.
Issue: #23164, #23161
