Some details in the readme were obsolete, some referring
to tools that are availbale for internal use of adacore only.
Installation manual now contains suffcient details,
no need to duplicate the same information here.
Follow-up of eng/toolchain/cuda#41
This commit enables on-board testing. To launch test with cross compilation
and execution on host:
* expose the environment (unless cuda toolbox is installed
in its standard location at /usr/local/cuda)
export CUDA_ROOT=<location of cuda root>
* make sure that $CUDA_ROOT/targets/aarch64-linux contains
copy of the same folder on the board
* load the ssh keys necessary for accessing the board
* launch tests
anod test gnat-cuda --target=aarch64-linux,,<board hostname or IP>
eng/toolchain/cuda#29
We don't support CUDA installations in /usr because the way we set
PATH then makes the system compiler in `/usr/bin/gcc`, which often
does not support Ada, shadow the compiler driver from GNAT. This
is also true for other directories that contain a installation of gcc.
This patch documents this restriction.
This patch adds some machinery to the build system to make it
possible to leverage a cross compiler for binding generation. It
also updates the documentation to remove the requirement of a native
compiler in the cross-compilation case.
Before this patch, the configuration file for the Sphinx
documentation had a workaround for an issue that's no longer present
with the version of Sphinx that comes with anod. This patch removes
that workaround.
The fact that the issue has been fixed upstream can be checked by
browsing the following GitHub issue:
https://github.com/readthedocs/readthedocs.org/issues/2569
The samples use the versions of cuda_api_host.gpr and cuda_api_device.gpr
that go through GPRinstall. The installing process removes the external
variable cuda_host, so setting it with the -X switch in Makefile.build
did not have an effect and could be misleading.
cuda_api_host.gpr was the only place where support for the CUDA_ROOT
environment variable was missing. This patch adds that support, and
with that it becomes possible to build the examples with a CUDA installation
that does not live in /usr/local/cuda.
Issue: eng/toolchain/cuda#32
