Related articles |
---|
Few Doubts about register allocation via Graph colouring [Briggs thesi shafitvm@gmail.com (Mohamed Shafi) (2007-10-15) |
Re: Few Doubts about register allocation via Graph colouring [Briggs t preston.briggs@gmail.com (preston.briggs@gmail.com) (2007-10-15) |
Re: Few Doubts about register allocation via Graph colouring [Briggs t shafitvm@gmail.com (shafi) (2007-10-18) |
Re: Few Doubts about register allocation via Graph colouring [Briggs t preston.briggs@gmail.com (preston.briggs@gmail.com) (2007-10-18) |
Re: Few Doubts about register allocation via Graph colouring [Briggs t bergner@vnet.ibm.com (Peter Bergner) (2007-10-26) |
From: | shafi <shafitvm@gmail.com> |
Newsgroups: | comp.compilers |
Date: | Thu, 18 Oct 2007 13:20:18 -0000 |
Organization: | Compilers Central |
References: | 07-10-04707-10-050 |
Keywords: | registers |
Posted-Date: | 18 Oct 2007 09:51:14 EDT |
> > 4. In the simplify phase Briggs has mentioned the following :
> > "In preparation for simplify, we classfiy each node based on its
> > degree and its spill cost. All the nodes of degree < k are placed in a
> > set 'low'. Nodes of degree >= k and finite spill cost are placed in a
> > set 'high'. nodes of the high degree and infinite spill cost are not
> > included in either set."
> > Position of the live ranges in the stack after they are pushed during
> > the simplify phase does play a important role in picking a colour.
> > Since the live ranges with infinite spill cost has to get a register,
> > they should be pushed on to the stack at the end so that they get a
> > register in the beginning itself. Am i right?
>
> Not exactly. A live range with infinite spill cost and low degree
> should be placed in set "low", where they will be removed from the
> graph and pushed on the stack early.
>
> If you just follow the description, things will end up in the right
> place; you don't have to add any more special cases.
Sir thanks for your reply.
I was asking about live ranges with infinite spill cost and "high"
degree. The pseudo code the follows the above description in your
thesis doesn't mention about such live ranges. But still they have to
be pushed. So again since they have to get a register should they be
pushed on to the stack at the end?
Regards,
Shafi
Return to the
comp.compilers page.
Search the
comp.compilers archives again.