- Red Hat Enterprise Linux (RHEL)* or Rocky Linux* 8
- SUSE Linux Enterprise Server (SLES)* or openSUSE Leap* 15
- Ubuntu* 18.04, 20.04, or 22.04 LTS
- GCC 7.4.0 and higher
- CMake 3.10 and higher
- Ninja 1.8.2 and higher
- Git
- libelf
- zlib (only needed for simulation flow)
- Windows* 10 (64 bit)
- Windows Server* 2012, 2016, or 2019
- Microsoft Visual C++ (MSVC)* 2017 and higher
- CMake 3.10 and higher
- Ninja 1.8.2 and higher
- Git
- Cygwin
- libelf (install with
scripts/install_libelf.ps1
) - zlib (only needed for simulation flow, install with
scripts/install_zlib.ps1
)
Perform these steps to build the runtime:
-
Clone and change to the runtime repository.
git clone https://github.com/intel/fpga-runtime-for-opencl cd fpga-runtime-for-opencl
-
Switch to the latest release version.
git checkout v2024.0
-
Create and change to the runtime build directory.
mkdir build cd build
-
Generate the Ninja build files using the
cmake
command.cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=../install
-
Build the runtime.
ninja
-
Install the runtime to the runtime installation directory.
ninja install
-
Build without debugging symbols and with optimization.
cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=../install
-
Build with debugging symbols and without optimization.
cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Debug -DCMAKE_INSTALL_PREFIX=../install
-
Use GCC as the compiler instead of the system default (for Linux only).
CC=gcc CXX=g++ cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=../install
-
Specify installation paths to libelf and zlib (for Windows only).
cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH=C:\libelf;C:\zlib -DCMAKE_INSTALL_PREFIX=../install
Alternatively, specify header and library directories explicitly.
cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INCLUDE_PATH=C:\libelf\include;C:\zlib\include -DCMAKE_LIBRARY_PATH=C:\libelf\lib;C:\zlib\lib -DCMAKE_INSTALL_PREFIX=../install
To run unit tests, execute the following command in the runtime build directory:
ctest -V
-
To run unit tests, you need the
aoc
executable from the IntelⓇ FPGA SDK for OpenCL™.To set the path to
aoc
, change to the IntelⓇ FPGA SDK for OpenCL™ installation directory and source the initialization script.-
On Linux:
source init_opencl.sh
-
On Windows:
call init_opencl.bat
-
-
On Linux,
aoc
requires thelibtinfo.so.5
library, which you can install using one of the following OS-specific commands:-
Red Hat Enterprise Linux (RHEL)* or Rocky Linux* 8:
sudo yum install ncurses-compat-libs-6.1
-
SUSE Linux Enterprise Server (SLES)* or openSUSE Leap* 15:
sudo zypper install libncurses5
-
Ubuntu* 18.04 or 20.04 LTS:
sudo apt install libtinfo5
-
-
On Windows, you need to set the paths to the libelf and (optionally) zlib libraries.
set PATH=C:\libelf\lib;C:\zlib\lib;%PATH%
-
Running unit tests is not supported for ARM cross compilation.
-
To run unit tests individually, first set the board package directory:
export AOCL_BOARD_PACKAGE_ROOT="$(git rev-parse --show-toplevel)/test/board/a10_ref"
Switch to the directory containing the test executable:
cd build/test
Run the tests whose group and name contain the given substrings:
./acl_test -v -g acl_mem -n create_buffer
Perform these steps if you have the IntelⓇ FPGA SDK for OpenCL™ or IntelⓇ FPGA RTE for OpenCL™ installed on your system:
-
In the IntelⓇ FPGA SDK for OpenCL™ installation directory, source the initialization script (see Running Unit Tests).
-
Update the
/etc/OpenCL/vendors/Altera.icd
file with either the filenamelibalteracl.so
ifLD_LIBRARY_PATH
contains the full path of the runtime installation directory; otherwise, the full path oflibalteracl.so
itself. -
Compile your OpenCL host program with the OpenCL header files included from the runtime installation directory.
-Ifpga-runtime-for-opencl/install/include
-
Link your OpenCL host program with the
libOpenCL.so
library from the IntelⓇ FPGA SDK for OpenCL™ installation directory.-L$INTELFPGAOCLSDKROOT/host/linux64/lib -lOpenCL
-
Run your OpenCL host program.
Perform these steps if you do not have the IntelⓇ FPGA SDK for OpenCL™ or IntelⓇ FPGA RTE for OpenCL™ installed on your system:
-
Download, build, and install the OpenCL ICD Loader.
-
Create the
/etc/OpenCL/vendors/Altera.icd
file with either the filenamelibalteracl.so
ifLD_LIBRARY_PATH
contains the full path of the runtime installation directory; otherwise, the full path oflibalteracl.so
itself. -
Compile your OpenCL host program with the OpenCL header files included from the runtime installation directory.
-Ifpga-runtime-for-opencl/install/include
-
Link your OpenCL host program with the
libOpenCL.so
library from the OpenCL ICD Loader.-lOpenCL
-
Run your OpenCL host program.
Environment Variable Title | Description |
---|---|
ACL_DEBUG | Set this variable to a value of 1 to 4 to output general-purpose debug messages from the OpenCL Runtime. |
ACL_HAL_DEBUG | Set this variable to a value of 1 to 5 to increase debug output from the hardware abstraction layer (HAL), which interfaces directly with the MMD layer. |
ACL_CONTEXT_CALLBACK_DEBUG | Set this variable to a value of 1 to produce error text for errors that occur in the OpenCL context. This is error text that would be sent upstream through a callback function provided with clCreateContext . |
- When setting the environment variable
OCL_ICD_FILENAMES
for debugging, ensure it doesn't resolve to anylibalteracl.so
other than the one specified in the/etc/OpenCL/vendors/Altera.icd
file. Mismatches may lead to multiple runtime library instances being linked into the program, causing undefined behaviour.