Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/beehive-lab/levelzero-jni
Intel LevelZero JNI library for TornadoVM
https://github.com/beehive-lab/levelzero-jni
gpu-library heterogeneous-parallel-programming intel-gpu java jni-library level-zero oneapi spirv
Last synced: 5 days ago
JSON representation
Intel LevelZero JNI library for TornadoVM
- Host: GitHub
- URL: https://github.com/beehive-lab/levelzero-jni
- Owner: beehive-lab
- License: other
- Created: 2021-02-19T12:36:12.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2024-09-17T07:20:25.000Z (about 2 months ago)
- Last Synced: 2024-09-17T09:59:31.546Z (about 2 months ago)
- Topics: gpu-library, heterogeneous-parallel-programming, intel-gpu, java, jni-library, level-zero, oneapi, spirv
- Language: Java
- Homepage:
- Size: 333 KB
- Stars: 11
- Watchers: 8
- Forks: 6
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- License: LICENSE
Awesome Lists containing this project
- awesome-oneapi - levelzero-jni - Intel LevelZero JNI library for TornadoVM. This project is a Java Native Interface (JNI) binding for Intel's Level Zero. This library is as designed to be as closed as possible to the LevelZero API for C++. (Table of Contents / Tools and Development)
README
# Beehive LevelZero JNI
Baremetal GPU and FPGA programming for Java using Intel's [Level Zero API](https://spec.oneapi.io/level-zero/latest/index.html). This project is a Java Native Interface (JNI) binding for Intel's Level Zero. This library is designed to be as closed as possible to the Level Zero API for C++. Subset of Level Zero 1.4.0 supported (Level Zero May 2022 version)
This project contains the logic needed to dispatch TornadoVM applications on Intel ARC and Intel integrated GPUs.
It is, by no means, a complete port of the Level Zero API, but external contributions are welcome.
## Compilation & configuration of the JNI Level Zero API
### 1) Compile Native Library: Level Zero API
#### Linux (Tested for Fedora 35-40 and Ubuntu 22.X-24.X)
```bash
git clone https://github.com/oneapi-src/level-zero.git
cd level-zero
mkdir build
cd build
cmake ..
cmake --build . --config Release
cmake --build . --config Release --target package
```
#### Windows 10/11
Tested Configurations:
- Lenovo IdeaPad Gaming 3 15IHU6
- Dell XPS 8950
- Windows 11
- VS Community 2022
+ components: C++, Git, Spectre mitigated libraries
- CMake 3.26.3, Maven 3.9.1, JDK 21+
Run commands in _x64 Native Tools Command Prompt for VS 2022_.
```cmd
git clone https://github.com/oneapi-src/level-zero
cd level-zero
md build
cd build
cmake ..
cmake --build . --config Release
rem check
.\bin\Release\zello_world.exe
```
Note: If `zello_world.exe` fails, search for existing Level Zero API DLLs (file names start with `ze_`, e.g. `ze_tracing_layer.dll`) in `c:\windows\system32` and move them to another folder.
### 2) Compile Level Zero Java JNI native code
Set the paths to the directory of Level Zero installation. Here are examples:
#### Linux
```bash
scl enable devtoolset-9 bash # << Only for CentOS
git clone https://github.com/beehive-lab/levelzero-jni
export ZE_SHARED_LOADER="/build/lib/libze_loader.so"
export CPLUS_INCLUDE_PATH=/include:$CPLUS_INCLUDE_PATH
export C_INCLUDE_PATH=/include:$CPLUS_INCLUDE_PATH
cd levelzero-jni/levelZeroLib
mkdir build
cd build
cmake ..
make
```
#### Windows 10/11
Note: Run commands in _x64 Native Tools Command Prompt for VS 2022_.
```cmd
git clone https://github.com/beehive-lab/levelzero-jni
set ZE_SHARED_LOADER=%USERPROFILE%\lab\level-zero\build\lib\release\ze_loader.lib
set CPLUS_INCLUDE_PATH=%USERPROFILE%\lab\level-zero\include
set C_INCLUDE_PATH=%USERPROFILE%\lab\level-zero\include
rem add the folder with Intel's Level Zero API DLLs to PATH
set PATH=%USERPROFILE%\lab\level-zero\build\bin\release;%PATH%
cd levelzero-jni\levelZeroLib
md build
cd build
cmake ..
cmake --build . --config Release
```
#### Obtain a SPIR-V compiler (Linux and Windows)
In case you want to compile kernels from OpenCL C to SPIR-V and use the Level Zero API, you need to download the `llvm-spirv` compiler. The implementation we are currently using is [Intel LLVM](https://github.com/intel/llvm).
### 3) Compile and run a Java test
#### Linux
```bash
mvn clean package
./scripts/run.sh ## < This script compiles an OpenCL C program to SPIR-V using the llvm-spirv compiler (see 2.1)
```
The OpenCL C kernel provided for this example is as follows:
```c
__kernel void copyData(__global int* input, __global int* output) {
uint idx = get_global_id(0);
output[idx] = input[idx];
}
```
To compile from OpenCL C to SPIR-V:
```bash
clang -cc1 -triple spir copyData.cl -O0 -finclude-default-header -emit-llvm-bc -o copyData.bc
llvm-spirv copyData.bc -o copyData.spv
```
#### Windows
Note: Java programs that use levelzero-jni are based on DLLs, which are provided by Intel's Level Zero API. For these programs to find these DLLs, the PATH environment variable must contain the folder that contains the DLLs.
```cmd
mvn clean package
rem add the folder with Intel's Level Zero API DLLs to PATH
set PATH=%USERPROFILE%\lab\level-zero\build\bin\release;%PATH%
rem copyData.spv file expected in CWD
.\scripts\run.cmd
# more tests
.\scripts\copies.cmd
.\scripts\events.cmd
.\scripts\fences.cmd
.\scripts\kernelTimers.cmd
.\scripts\transferTimers.cmd
.\scripts\largeBuffers.cmd
```
## License
This project is developed at [The University of Manchester](https://www.manchester.ac.uk/), and it is fully open source under the [MIT](https://github.com/beehive-lab/levelzero-jni/blob/master/LICENSE) license.
## Acknowledgments
The work was partially funded by the EU Horizon 2020 [Elegant 957286](https://www.elegant-h2020.eu/) project, and [Intel Coorporation](https://www.intel.it/content/www/it/it/homepage.html).