This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: SPIR-V and GCC


On Thu, 2015-03-05 at 12:27 +0100, Xerxes RÃnby wrote:
> > On Wed, 2015-03-04 at 16:20 +0000, BogDan wrote:
> > > On Wednesday, March 4, 2015 5:22 PM, David Malcolm <dmalcolm@redhat.com> wrote:
> > > On Wed, 2015-03-04 at 14:36 +0000, BogDan wrote:
> > > > Hello folks,
> > > >
> > > >
> > > > Probably is a little bit too early, but I'd like to ask if there is
> > > > any chance to use gcc to produce SPIR-V [1].
> > > > It will be just great if we'll be able to write our shaders in e.g.
> > > > C/C++/(any language supported by gcc) and use GCC to compile them as
> > > > SPIR-V!
> > > > It will be fantastic to use only one compiler collection for CPU and
> > > > for GPU compute&graphics!
> > > >
> > > > Cheers,
> > > >
> > > > BogDan.
> > > > [1]https://www.khronos.org/registry/spir-v/
> > >
> > > I work on gcc, but in a former life I was a game developer [1], so this
> > > piqued my interest :)
> ...
> >
> > > SPIR-V it's an Intermediate Language binary format. First versions were
> > > based on LLVM. SPIR-V has nothing to do with the old SPIR implementations,
> > > but for some reason khronos decide to keep the name :)
> > > Thishttp://www.g-truc.net/post-0714.html  great article explans better
> > > what SPIR-V is.
> >
> > (nods); thanks.
> >
> > > (B) Are you thinking about this for primarly ahead-of-time
> > > compilation, or are you interested in just-in-time compilation to
> > > SPIR-V? I ask since I maintain the new "libgccjit" feature in GCC 5.
> > > One of the current assumptions in libgccjit is that host==target, but I
> > > hope to relax that for gcc 6 so that libgccjit could e.g. generate code
> > > for a GPU. Another gcc 6 possibility could be multi-target support for
> > > libgccjit, so that you can populate a gcc_jit_context with code, then
> > > have it generate machine code for both the CPU and for the GPU (mostly
> > > just thinking aloud here).
> >
> > > Even though most of the people will use it for ahead-of-time compilation,
> > > but IMHO it can be both!
> >
> > Another use-case that occurred to me when I was looking at the link you
> > posted above: using gcc for optimizing SPIR-V and compiling it e.g. to
> > CPU code.
> >
> > As of gcc 5, libgccjit can be hooked up to a pre-existing language
> > frontend, with the gcc backend emitting machine code for the CPU,
> > assuming that the frontend is license-compatible with libgccjit's
> > GPLv3-or-later.  See this example, which uses libgccjit to build a
> > compiler for brainf**k:
> >   https://gcc.gnu.org/onlinedocs/jit/intro/tutorial05.html
> >
> > So if there's a SPIR-V loader library that's license-compatible, you
> > could write code to inject SPIR-V into libgccjit, and call
> > gcc_jit_context_compile_to_file on it to optimize it and turn it into
> > machine code.
> >
> Johann Sorel has written a SPIR-v binary reader and writer with a compatible license (public domain).
> https://bitbucket.org/Eclesia/unlicense/src/tip/compiler/compiler-spir/src/main/java/un/language/spir/?at=default
> 
> To quote Johann Sorel: "it's public domain, copy what you need ;)"
> https://jogamp.org/bugzilla/show_bug.cgi?id=1140#c2
> 
> These SPIR-v classes is to my knowledge the first public SPIR-v reader and writer with a free software compatible license.
> You can use the GCC GCJ project and Johann Sorel's Java classes that gives you all SPIR-v spec constants in machine readable form
> and a SPIR-v binary parser and writer as a foundation to create your SPIR-v compiler & assembler.

Thanks.

FWIW, we don't yet have java bindings for libgccjit.  Writing java
bindings would be a non-trivial project, but probably not too hard for a
java expert (perhaps suitable as a GSoC project, if anyone's
interested?).

Dave


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]