[llvm-commits] Regalloc Refactoring

David A. Greene greened at obbligato.org
Wed May 9 16:54:43 PDT 2007

Hash: SHA1

Evan Cheng wrote:
> Hi,
> The patch seems pretty safe to me. The only thing I am not sure about is:


> It's not clear to me if this is the right way to model this pass. After
> all, this is not an analysis pass. Perhaps the right thing to do is to
> model this after the register allocator. Add a createRegisterCoalescer()
> to Passes.cpp which would allow us to choose the one we want. Then
> register the pass in LLVMTargetMachine.cpp. Chris? What do you say?

Yeah, I wasn't sure about that either.  I modeled it after the way that
out-of-ssa conversion is done since LinearScan assumes that coalescing
has been done.

I think your suggestion of createRegisterCoalescer() is the right
solution in the long run.  My goal here was to submit the first phase
of a larger refactoring effort that separated coalescing from
live interval analysis but tried not to do anything else dramatic.


Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org


More information about the llvm-commits mailing list