This is the mail archive of the
mailing list for the GCC project.
Re: LLVM/GCC Integration Proposal
- From: Tom Tromey <tromey at redhat dot com>
- To: Chris Lattner <clattner at apple dot com>
- Cc: Chris Lattner <sabre at nondot dot org>, GCC Mailing List <gcc at gcc dot gnu dot org>
- Date: 28 Nov 2005 13:43:18 -0700
- Subject: Re: LLVM/GCC Integration Proposal
- References: <37A3A6DD-711A-4856-ACC0-259FCD9841CF@apple.com>
- Reply-to: tromey at redhat dot com
>>>>> "Chris" == Chris Lattner <email@example.com> writes:
Chris> In this role, it provides a static optimizer, interprocedural link-
Chris> time optimizer, JIT support, and several other features.
I'm quite interested in the JIT aspect of LLVM, for gcj.
This would fill one of our major missing gaps.
However, it seems to me that for this to work well in practice, it
would mean that the code generators would have to be under a "runtime
friendly" license -- that is, GPL would probably not be appropriate.
This leads into somewhat ugly territory ... either requiring a "pure
LLVM" back end for JIT support, or relicensing RTL back ends to
Chris> The IR supports several features that are useful to various
Chris> communities, including true tail calls, accurate garbage collection,
Us gcj hackers would also like accurate GC support.
Java is fairly dynamic, as I'm sure you know. So, I'm much more
interested in the JITting possibilities than in link time
optimizations; the latter is cool and probably useful to embedded
users of gcj, but I'd imagine for all our recent binary compatibility
deployments we would just end up ignoring it.