On 17 Aug 2010, at 16:04, Kevan Hashemi wrote:
Keeping up with changes in the GCC IR definition is time-consuming.
As a side note: I just attended a presentation by Diego Novillo, who is at Google (previously at Cygnus and at Red Hat) and who works on GCC. He mentioned that one thing they are working on is a textual representation for GIMPLE (GCC's current intermediate representation). I explicitly asked and he confirmed that once that work is finished, it will be possible to use GCC in the same way as LLVM: as a generic code generation backend by processing textual GIMPLE generated by an independent frontend.
Of course, that is probably not that helpful in the current situation, because a) it will still take a while before such an interface becomes available b) even if it were targeted, the current stable GPC frontend is for GCC 3.x and GIMPLE was only introduced with GCC 4.0, so it would probably not save that much work
Jonas