NVIDIA opens up CUDA compiler for other languages

NVIDIA opens up CUDA compiler for other languages

[Dirk Craeynest]

Just read the following announcement:

<quote>
NVIDIA Opens Up CUDA Compiler

GPU maker NVIDIA is going to make its CUDA compiler runtime source
code, and internal representation format public, opening up the
technology for different programming languages and processor
architectures. [...]

Read More: <http://www.hpcwire.com/ct/uz5609392Biz12256180>"
</quote>

The full announcement says the LLVM compiler infrastructure will be
used as the vehicle for the public CUDA source code.  The main idea
seems to be to allow others to use the technology (among others)
with programming languages that NVIDIA is not pursuing themselves.

Interesting for Ada?

Dirk

Re: NVIDIA opens up CUDA compiler for other languages

[Adrian-Ken Rueegsegger]

Hi,

On 12/16/2011 08:05 AM, Dirk Craeynest wrote:
> Just read the following announcement:
> 
[snip]
> 
> The full announcement says the LLVM compiler infrastructure will be
> used as the vehicle for the public CUDA source code.  The main idea
> seems to be to allow others to use the technology (among others)
> with programming languages that NVIDIA is not pursuing themselves.
> 
> Interesting for Ada?

If I understood the article correctly this requires a working
LLVM-Frontend for Ada. I do not know what the status of the current
implementation [1] is but maybe somebody more knowledgeable about LLVM
with regards to Ada could comment on that and give their take?

Incidentally Reto Buerki and me have been working on an Ada-Binding for
CUDA. Those interested in using CUDA from Ada might want to check out
the project website [2] and the source repository [3].

The low-level thin-binding is complete and we added some abstractions on
top of that. Many ideas/goals were inspired by the excellent Python
binding pyCUDA [4]. We have not made a release yet since we are
currently working on the documentation. If you give CUDA/Ada a try we
would be happy if you would let us know what you think.

Cheers,
Adrian
--
[1] - http://llvm.org/docs/GCCFEBuildInstrs.html#ada
[2] - http://www.codelabs.ch/cuda-ada/
[3] - http://git.codelabs.ch/?p=cuda-ada.git
[4] - http://mathema.tician.de/software/pycuda

Re: NVIDIA opens up CUDA compiler for other languages

[jonathan]

On Dec 16, 1:43 pm, Adrian-Ken Rueegsegger <k...@codelabs.ch> wrote:

>
> If I understood the article correctly this requires a working
> LLVM-Frontend for Ada. I do not know what the status of the current
> implementation [1] is but maybe somebody more knowledgeable about LLVM
> with regards to Ada could comment on that and give their take?
>


I haven't had a chance to try it yet, but the new LLVM 3.0
has been out a few weeks, along with the new Ada front end

   http://dragonegg.llvm.org/

They say:

   Patching and building GCC is no longer required: the plugin should
   work with your system GCC (version 4.5 or 4.6; on Debian/Ubuntu
systems
   the gcc-4.5-plugin-dev or gcc-4.6-plugin-dev package is also
needed).

which sounds encouraging.

Jonathan

Re: NVIDIA opens up CUDA compiler for other languages

[Rugxulo]

Hi,
   I'm pretty sure I'm the wrong person to answer this, but oh
well.  ;-)

On Dec 16, 9:08 am, jonathan <johns...@googlemail.com> wrote:
> On Dec 16, 1:43 pm, Adrian-Ken Rueegsegger <k...@codelabs.ch> wrote:
>
> > If I understood the article correctly this requires a working
> > LLVM-Frontend for Ada. I do not know what the status of the current
> > implementation [1] is but maybe somebody more knowledgeable about LLVM
> > with regards to Ada could comment on that and give their take?
>
> I haven't had a chance to try it yet, but the new LLVM 3.0
> has been out a few weeks, along with the new Ada front end
>
>    http://dragonegg.llvm.org/
>
> They say:
>
>    Patching and building GCC is no longer required: the plugin should
>    work with your system GCC (version 4.5 or 4.6; on Debian/Ubuntu
> systems
>    the gcc-4.5-plugin-dev or gcc-4.6-plugin-dev package is also
> needed).
>
> which sounds encouraging.

Clang does not have an Ada frontend. It's a "C language"-based
compiler only, e.g. C / C++ / Objective C / Objective C++. LLVM is the
backend, hence the (now deprecated, no longer updated, not available
in LLVM 3.0) LLVM-GCC (GCC 4.2) compiler was needed to compile Ada
source code while targeting the LLVM backend.

DragonEgg is a plugin for GCC proper (specifically, 4.5.x or 4.6.x)
which targets the LLVM backend. Since GCC (only, and not Clang) has an
Ada frontend / compiler, you must use that.