llvm dev - Oct 2008 - [LLVMdev] XMOS using LLVM

Hi,

I'm a compiler engineer at XMOS (http://www.xmos.com) and in the last
few months I've been working on porting LLVM to target our XS1-G4 chip.
I thought it may be of interest to the list to find out how we are using
of LLVM.

The XS1-G4 has four processors and 32 hardware threads. It has been
designed to be highly responsive to I/O events allowing many tasks
normally be done by hardware to instead by carried out in software. It
has hardware support for multi-threading, inter thread communication and
efficient I/O operations. More information is available on our website.

The latest release of our software tool chain now contains a C compiler
based on llvm-gcc and LLVM 2.3. Features supported by our backend
include inline assembly, dwarf debug support, 64 bit long long support
and floating point using LLVM's softfloat. The tools can be downloaded
for free after registration with source code for our modified LLVM /
llvm-gcc. At the moment we are using LLVM for C compilation only.
Parallelism and I/O are expressed using a C like language called XC for
which we have a separate bespoke compiler. We are considering adding
support for some of the more novel features of our architecture into
LLVM and maybe merging these two compilers in future.

I have found LLVM a joy to use. It has taken just over 3 months
of work to get to this point from scratch (this also includes porting a
C standard library). LLVM clearly benefits from its modularity - the
code is easy to understand and the interfaces are well designed. We have
also been pleased with the results (efficiency of code produced and
compilation time).

I would like to gauge is how much interest there would be in having our
backend make its way back upstream. If nothing else it would provide a
good example of a fairly complete LLVM backend targeting a nice, simple
RISC instruction set: I know I found the existing backends very helpful
for reference during development. I've read through the developer policy
but wasn't sure how your copyright assignment policy works (would we
need to sign something to contribute?). Where would we go from here?

Richard Osborne | XMOS
http://www.xmos.com

Hi Richard,

This is very cool stuff. Thank for you sharing your experience with  
all of us.

I would encourage you to contribute your target back to the  
repository. Assuming you have the right tests, this will ensure all  
future works on the top of tree will not intentionally break it. And  
as you have stated, the community can benefit from having another  
interesting target to learn from.

Evan

On Oct 14, 2008, at 12:34 PM, Richard Osborne wrote:
> Hi,
>
> I'm a compiler engineer at XMOS (http://www.xmos.com) and in the last
> few months I've been working on porting LLVM to target our XS1-G4  
> chip.
> I thought it may be of interest to the list to find out how we are  
> using
> of LLVM.
>
> The XS1-G4 has four processors and 32 hardware threads. It has been
> designed to be highly responsive to I/O events allowing many tasks
> normally be done by hardware to instead by carried out in software. It
> has hardware support for multi-threading, inter thread communication  
> and
> efficient I/O operations. More information is available on our  
> website.
>
> The latest release of our software tool chain now contains a C  
> compiler
> based on llvm-gcc and LLVM 2.3. Features supported by our backend
> include inline assembly, dwarf debug support, 64 bit long long support
> and floating point using LLVM's softfloat. The tools can be downloaded
> for free after registration with source code for our modified LLVM /
> llvm-gcc. At the moment we are using LLVM for C compilation only.
> Parallelism and I/O are expressed using a C like language called XC  
> for
> which we have a separate bespoke compiler. We are considering adding
> support for some of the more novel features of our architecture into
> LLVM and maybe merging these two compilers in future.
>
> I have found LLVM a joy to use. It has taken just over 3 months
> of work to get to this point from scratch (this also includes  
> porting a
> C standard library). LLVM clearly benefits from its modularity - the
> code is easy to understand and the interfaces are well designed. We  
> have
> also been pleased with the results (efficiency of code produced and
> compilation time).
>
> I would like to gauge is how much interest there would be in having  
> our
> backend make its way back upstream. If nothing else it would provide a
> good example of a fairly complete LLVM backend targeting a nice,  
> simple
> RISC instruction set: I know I found the existing backends very  
> helpful
> for reference during development. I've read through the developer  
> policy
> but wasn't sure how your copyright assignment policy works (would we
> need to sign something to contribute?). Where would we go from here?
>
> Richard Osborne | XMOS
> http://www.xmos.com
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

On Oct 14, 2008, at 12:34 PM, Richard Osborne wrote:
> I would like to gauge is how much interest there would be in having  
> our
> backend make its way back upstream.
I think it would be good to have more ports.
> I've read through the developer policy but wasn't sure how your  
> copyright assignment policy works (would we need to sign something  
> to contribute?). Where would we go from here?
My take, have your lawyers and VP of engineering review it, and it you  
want to assign it over to the llvm project, just email the patch to  
the development list.  To do this, you have to have the right to  
assign the work.