This is actually a big CL, which I'm not sure we can get into landable state
soonish. I suspect that we could proceed a lot faster if we split the work
into
manageable parts, something along these lines:
1.) Do the minimal amount of work required to make the "greedy allocator
bot"
green (i.e. passing all our current tests).
2.) Implement the basic algorithm (as mentioned in
http://blog.llvm.org/2011/09/greedy-register-allocation-in-llvm-30.html)
without
any fancy heuristics, but with the interface introduced by LLVM to make the
splitting/spilling decisions.
3.) Stabilize the basic algorithm and do measurements against LS.
4.) Based on that, split the basic allocator from LS and use appropriate
data
structures instead of trying to share all datastructures with LS.
5.) Start implementing the greedy heuristics on top of the basic allocator,
just
like LLVM did.
6.) Weaponize.
WDYT?
https://codereview.chromium.org/1157663007/
--
--
v8-dev mailing list
v8-dev@googlegroups.com
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to v8-dev+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
