llvm dev - Oct 2011 - [LLVMdev] [cfe-dev] ANN: libclc (OpenCL C library implementation)

Hi,
> libclc: http://www.pcc.me.uk/~peter/libclc/
> pocl: https://launchpad.net/pocl
> clover: http://cgit.freedesktop.org/~steckdenis/clover/
I have pushed our implementation on GitHub:

https://github.com/speziale-ettore/OpenCRun

I have focused on desiging a modular system, because OpenCRun is
intended to be used for research purpose -- not many people, no too much
time, so clean design is not an option. For the same reason, I have
tried to re-use as much as possible what LLVM/CLANG offers: compiler,
jit, build system, diagnostic routines, testing infrastructure, ...

Currently I have implemented an OpenCL backend for i386 and X86_64 CPUs.
I have coded work-item stacks by hand because after some sintetic
benchmarks I have observed that both ucontext and setjmp/longjmp contain
some extra overheads.

Code is not fully documented -- sorry -- and some mandatory features are
still missing (e.g. local variables).

Looking at the code, you can observe that there are a lot of LLVM
passes. I have implemented some features through passed because I want
to keep the code as much modular as possible.

Best regards,
speziale.ettore at gmail.com

PS-1: OpenCRun requires some extra metadata to be generated by clang. In
sources root directory there is a clang patch to add support for
metadata generation

PS-2: some parts are very bad-coded. I know, but I haven't had yet time
to do some refactoring

On Fri, Oct 21, 2011 at 9:48 AM, Speziale Ettore
<speziale.ettore at gmail.com> wrote:
> Currently I have implemented an OpenCL backend for i386 and X86_64 CPUs.
> I have coded work-item stacks by hand because after some sintetic
> benchmarks I have observed that both ucontext and setjmp/longjmp contain
> some extra overheads.
If you have not already seen it, you (and anyone else working on
OpenCL runtimes) might be interested in this paper from AMD:
http://dl.acm.org/citation.cfm?id=1854302
In particular, Section 4 describes the implementation approach for
their x86 OpenCL runtime and mentions a number of optimizations they
applied to things like the work-item stack.

- Michael

Hi,
> If you have not already seen it, you (and anyone else working on
> OpenCL runtimes) might be interested in this paper from AMD:
> http://dl.acm.org/citation.cfm?id=1854302
> In particular, Section 4 describes the implementation approach for
> their x86 OpenCL runtime and mentions a number of optimizations they
> applied to things like the work-item stack.
I know, but I hadn't time to apply all the optimizations -- I have only
coded the most trivial.

Best regards,
speziale.ettore at gmail.com