https://github.com/pocl/pocl
pocl - Portable Computing Language
https://github.com/pocl/pocl
heterogeneous-parallel-programming opencl
Last synced: 3 months ago
JSON representation
pocl - Portable Computing Language
- Host: GitHub
- URL: https://github.com/pocl/pocl
- Owner: pocl
- License: mit
- Created: 2013-08-22T09:26:12.000Z (almost 12 years ago)
- Default Branch: main
- Last Pushed: 2024-04-19T09:10:21.000Z (about 1 year ago)
- Last Synced: 2024-04-19T14:04:51.068Z (about 1 year ago)
- Topics: heterogeneous-parallel-programming, opencl
- Language: C
- Homepage: http://portablecl.org
- Size: 55.5 MB
- Stars: 867
- Watchers: 78
- Forks: 242
- Open Issues: 103
-
Metadata Files:
- Readme: README.PPC64le
- Changelog: CHANGES
- License: COPYING
Awesome Lists containing this project
README
PoCL is used to provide OpenCL on IBM AC922 computers
featuring IBM Power9 processors and Nvidia Tesla V100 GPU
interconnected with NVlink v2 (up to 72 GByte/s).
This has been tested under debian_11 and Ubuntu_20.04.
Officially, Nvidia does not support OpenCL on this platform
and the driver they are shipping is lacking the compiler part.
## Building tricks (as of 04/2023):
The PPC64le features 128-bit vector unit (Altivec/VSX) which
are easily confused by the C++ compiler with the C++ vector
instruction when using the compile option `-std=c++XX`.
The corresponding code usually fails compiling.
The trick is to pass the option `-std=gnu++XX`.
For example, when configuring pocl:
```
cmake .. -DLLVM_CXXFLAGS="-std=gnu++14 ..." -DENABLE_CUDA=ON
```
The full list of options for the CXXFLAGS is obtained with:
```
llvm-config --cxxflags|sed -e "s/std=c/std=gnu/"
```
Later on, the build continues with `make` ...
Note the CUDA option to enable the GPU support on those computers.