Portable Computing Language - an OpenCL implementation
Portable Computing Language (pocl) aims to become a MIT-licensed open source
implementation of the OpenCL standard which can be easily adapted for new targets
and devices, both for homogeneous CPU and heterogenous GPUs/accelerators.
pocl uses Clang as an OpenCL C frontend and LLVM for the kernel compiler
implementation, and as a portability layer. Thus, if your desired target
has an LLVM backend, it should be able to get OpenCL support easily by using pocl.
The goal is to accomplish improved performance portability using a kernel
compiler that can generate multi-work-item work-group functions that exploit
various types of parallel hardware resources: VLIW, superscalar, SIMD, SIMT,
multicore, multithread ...
Additional purpose of the project is to serve as a research platform for
issues in parallel programming on heterogeneous platforms.
- Links to science / pocl
- Has a link diff
- Download package
-
Checkout Package
osc -A https://api.opensuse.org checkout home:mnhauke:opencl:stable/pocl && cd $_ - Create Badge
Source Files
| Filename | Size | Changed |
|---|---|---|
| _link | 0000000186 186 Bytes | |
| pocl-1.6.tar.gz | 0001595212 1.52 MB | |
| project.diff | 0000004153 4.06 KB |
Revision 2 (latest revision is 5)
Martin Hauke (mnhauke)
committed
(revision 2)
- Update to version 1.6
* Support for LLVM 11.
* CUDA kernels using constant __local blocks are now ABI
incompatible with previous release. Users need to delete their
pocl cache.
* Improved debugging of OpenCL code with CPU driver.
* Improved the PTX code generation for __local blocks.
* Improved handling of command queue barriers
* Fix LLVM loop vectorizing remarks printing
(POCL_VECTORIZER_REMARKS=1).
* Fix an issue in which the loop vectorizer produced code with
invalid memory reads (issue #757).
* Fix compilation error when CMake option SINGLE_LLVM_LIB is
set to OFF.
* Fix wrongly output dlerror (Undefined symbol) after dlopen,
caused by a previous libdl call in an ICD loader
* [CPU] safety margin of pocl's CPU driver local memory
allocation has been reduced to a much more reasonable value
* [CPU] buffer size for OpenCL printf is now configurable with
PRINTF_BUFFER_SIZE CMake variable
* [CPU] local memory size reported is now the size of last level
of non-shared data cache (usually L1 or L2 depending on CPU),
if hwloc can determine it.
- Update patch link_against_libclang-cpp_so.patch
Comments 0