[LLVMdev] "Graphite" for llvm
gfursin at gmail.com
Mon Jan 4 16:24:13 PST 2010
Happy New Year!
Just wanted to mention that since cTuning community is interested in
performance/code size/power tuning using empirical feedback-directed techniques
using different compilers including GCC and LLVM, so just wanted to mention that
we would be interested to add support to fine-grain optimizations from GRAPHITE
to Interactive Compilation Interface at some point at the beginning
of this year to be able to use cTuning optimization framework directly
and share optimization cases with the community.
Will keep in touch,
> -----Original Message-----
> From: gcc-graphite at googlegroups.com [mailto:gcc-graphite at googlegroups.com] On Behalf Of Albert
> Sent: Tuesday, December 29, 2009 1:12 PM
> To: Tobias Grosser
> Cc: ether; LLVM Developers Mailing List; GCC GRAPHITE; loopo at infosun.fim.uni-passau.de
> Subject: Re: [LLVMdev] "Graphite" for llvm
> Tobias Grosser wrote:
> > The way to go is the scoplib format (propably extended by quantified
> > variables). This format could be extracted from graphite easily and
> > could also be created in LLVM.
> > What we need to get back into LLVM is only the new optimized schedule
> > described e.g. as cloog like scattering functions. These can be parsed
> > easily. The real code generation could be done internally, so it is not
> > necessary to parse the generated from external tools.
> By the way, Konrad Trifunovic is interested (for his own research) to
> improve on the current read/write capabilities of Graphite to process
> the full scoplib format instead. Dealing with iteration domain and
> schedule/scattering is easy, but to process array subscripts will
> require more work. We will keep you informed of the progresses on the
> Graphite mailing list, but collaboration outside the Graphities is welcome.
More information about the llvm-dev