Switching XRT development work from P4 to Git can be done without much downtime provided you use a few scripts we have created:
build.sh
build script that builds XRT for both Debug and Release profiles.run.sh
loader script that sets up environment assuming XRT was built withbuild.sh
.board.sh
harvests spriteUNIT_HW
test cases and runs board tests.
Make sure you build XRT on a Centos7.4+ or Ubuntu16.04 or Ubuntu18.04 host.
It is probably safest if you keep your Git clone of XRT on a network mounted drive that can be accessed from different hosts. One advantage is that you can have your editor run on a host that is not used for board testing, since you don't really want host/driver crashes to leave your unsaved edits in limbo.
git clone https://github.com/Xilinx/XRT.git cd XRT/build ./build.sh
build.sh
script builds for both Debug and Release profiles. It is
necessary to use the build script if you intend to use the loader
script run.sh
and the board testing script board.sh
.
For the normal development flow, it is not necessary to build RPM or DEB packages. The loader and test scripts both work by setting the environment to point at the binaries created by the build script.
To run your locally built XRT with a sample host.exe
and
kernel.xclbin
, simply prepend your command line invocation with
XRT/build/run.sh
<path>/XRT/build/run.sh ./host.exe kernel.xclbin
By default the run.sh
script uses the binaries from the Release
profile. In order run with the binaries from Debug profile use -dbg
flag; this way you can even start your favorite debugger by prefixing its
invocation with run.sh -dbg
<path>/XRT/build/run.sh -dbg emacs
After making changes to XRT in your Git clone, rebuild with
build.sh
as explained above, then run a full set of board tests
using the board.sh
script. For example:
mkdir tests cd tests <path>/XRT/build/board.sh -board vcu1525 -sync
The -sync
option tells the script to rsync tests from the latest
nightly sprite area. Without the -sync
option, the board script will
run all tests that were previously synced into the current directory.
While tests run a file named results.all
will list the test with
PASS
/FAIL
keyword. This file is appended (not removed
between runs). A complete run should take 5-10 mins for approximately
70 tests.
We use GTest to do unit testing. The GTest package is installed by
running XRT/src/runtime_src/tools/scripts/xrtdeps.sh
.
The GTest package on CentOS/RHEL 7.5 provides the GTest libraries here:
/usr/lib64/libgtest.so
/usr/lib64/libgtest_main.so
In recent versions of Ubuntu, the GTest libgtest-dev
package
provides the compiled libraries in
/usr/lib/x86_64-linux-gnu/libgtest.a
/usr/lib/x86_64-linux-gnu/libgtest_main.a
However, the GTest package on Ubuntu up to 18.04 provides source only! So, to use GTest on older Ubuntu versions, use:
cd /usr/src/gtest sudo cmake CMakeLists.txt sudo make cd /usr/lib sudo ln -s /usr/src/gtest/libgtest.a sudo ln -s /usr/src/gtest/libgtest_main.a # Validate: ls *gtest*
This will add GTest static library symbolic links here:
/usr/lib/libgtest.a
/usr/lib/libgtest_main.a
CMake will handle linking, finding etc. for you.
To add GTest support to a CMakeLists.txt
use the following, and this is using
an example executable called xclbintest
:
find_package(GTest) if (GTEST_FOUND) enable_testing() message (STATUS "GTest include dirs: '${GTEST_INCLUDE_DIRS}'") include_directories(${GTEST_INCLUDE_DIRS}) add_executable(xclbintest unittests/main.cpp unittests/test.cpp) message (STATUS "GTest libraries: '${GTEST_BOTH_LIBRARIES}'") target_link_libraries(xclbintest ${GTEST_BOTH_LIBRARIES} pthread) else() message (STATUS "GTest was not found, skipping generation of test executables") endif()