The Intel Fortran compiler supports an extension that allows conditional
compilation based on preprocessor definitions specified on the command
line even when not preprocessing.
Fixes: #19664
dca9c33abc Tests: Remove old IPO test
c856d4556b bindexplib: supporting llvm bitcode formats using llvm-nm
079b8e2916 Clang: prefer lld-link over link.exe
6e3655db2c Clang: add LTO support for GNU-command line clang on windows
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !3527
25f29b9741 cuda: Adapted tests to work with modified cuda toolset
ee86770a3f cuda: Added docs for extended cuda toolset
0ad180d712 cuda: Extend cuda compiler detection to work with custom cuda path
55b0532128 cuda: Extend vs10 target generator to use custom cuda path
df0247a371 cuda: Extend toolset argument to accept path
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !3713
7828577065 FindBoost: Add support for Boost 1.71
cf29fa18c8 FindBoost: Unwrap compatibility INTERFACE targets for legacy variables
d7df3cd73f FindBoost: Remove incorrect 1.70 timer dependency
8ff43de1a5 FindBoost: Simplify conditional block for last known version
Acked-by: Kitware Robot <kwrobot@kitware.com>
Acked-by: Dennis Klein <d.klein@gsi.de>
Merge-request: !3763
Since commit 0dd6772a89 (FindBoost: Add legacy variables and targets for
compatibility, 2019-06-14, v3.15.0-rc2~3^2~1) we extract information
from imported targets provided by upstream `BoostConfig.cmake` files.
However, upstream Boost 1.71 provides some imported targets only for
compatibility. They are just INTERFACE libraries that wrap around the
real target named by `INTERFACE_LINK_LIBRARIES`. Unwrap this layer so
we can extract the real imported location.
Fixes: #19656
Running `Utilities/Scripts/BoostScanDeps.cmake` with the Boost 1.70.0
sources shows that the `timer` component no longer depends on `system`.
This is consistent with upstream Boost Timer commit `43eecbd071`.
For the multi-configuration generators one can specify the list
of configurations to include in the package.
E.g. having a project, where debug libraries have a suffix to
distinct them from the release builds, one can build the package
containing both `Debug` and `Release` binaries.
When running `pkg_search_module`, it may be useful to get the matching
module name in order to run `pkg_get_variable`.
`pkg_search_module` now defines `<prefix>_MODULE_NAME` which contains
the first matching module name.
Fixes: #19648
It is currently presumed that the linker will generate
a space separated line. This is not the case for some
compilers such as IBM XL where it can output space or
comma separated. This is particularly apparent when IBM XL
is used as the host compiler for CUDA as it generates a comma
separated link line.
To make using the generated STGZ easier, require a specific answer to
accepting the license terms. Since more moves down one line when '\n'
is entered a user may hold enter to paginate through the document. This
change prevents the user from accidentally refusing the license terms by
holding enter for too long and having to start over.
Beginning with the 19.06 release of the Cray Programming Environment, the
default linking mode on XC Cray systems is dynamic instead of static.
This updates the CrayPrgEnv compiler wrappers to detect the PE version
being used and behave accordingly.
In commit e374b9f1eb (FindMPI: Restore MPI_<LANG>_COMPILE_FLAGS as a
command-line string, 2018-09-10, v3.12.3~6^2) we tried to restore the
compile flags but accidentally inverted the arguments to `list(JOIN)`
causing both `MPI_<LANG>_COMPILE_FLAGS` and `MPI_<LANG>_COMPILE_OPTIONS`
to be empty.
Issue: #18349
The previous code did not count the number of virtual processors across
physical processors on SunOS.
Can handle the following situations:
```
$ uname -a
SunOS solaris 5.11 11.4.0.15.0 i86pc i386 i86pc
$ psrinfo -p -v
The physical processor has 1 virtual processor (0)
x86 (GenuineIntel 306D2 family 6 model 61 step 2 clock 3492 MHz)
Intel Core Processor (Broadwell)
The physical processor has 1 virtual processor (1)
x86 (GenuineIntel 306D2 family 6 model 61 step 2 clock 3492 MHz)
Intel Core Processor (Broadwell)
The physical processor has 1 virtual processor (2)
x86 (GenuineIntel 306D2 family 6 model 61 step 2 clock 3492 MHz)
Intel Core Processor (Broadwell)
The physical processor has 1 virtual processor (3)
x86 (GenuineIntel 306D2 family 6 model 61 step 2 clock 3492 MHz)
Intel Core Processor (Broadwell)
```
```
$ uname -a
SunOS sol11x86 5.11 11.0 i86pc i386 i86pc
$ psrinfo -p -v
The physical processor has 2 virtual processors (0 1)
x86 (GenuineIntel 50650 family 6 model 85 step 0 clock 2000 MHz)
Intel(r) Xeon(r) Gold 6138 CPU @ 2.00GHz
```
```
$ uname -a
SunOS sol11 5.11 11.0 sun4v sparc sun4v
$ psrinfo -p -v
The physical processor has 2 cores and 16 virtual processors (0-15)
The core has 8 virtual processors (0-7)
The core has 8 virtual processors (8-15)
SPARC-T4 (chipid 0, clock 2848 MHz)
```
Backslashes in `CUDA_NVCC_INCLUDE_DIRS` cause syntax errors because it
is written to a generated `.cmake` script as a string literal. Use a
bracket argument to hold the expanded value as we already do for
`CUDA_NVCC_COMPILE_DEFINITIONS`. Then explicitly convert backslashes to
forward slashes.
f4fc0667ae FindOpenACC: Provide import targets and OpenACC_<lang>_OPTIONS variable
9460501ad7 FindOpenACC: Provide a Fortran snippet that compiles with gfortran
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !3670
Previously the FindOpenACC module had issues where the contents of
OpenACC_<lang>_FLAGS could not be used with target_compile_options
when it contained multiple compiler flags.
