[prev in list] [next in list] [prev in thread] [next in thread] 

List:       pcc-list
Subject:    More about liveness of regs
From:       Anders Magnusson <ragge () dc ! ltu ! se>
Date:       2007-11-06 15:55:30
Message-ID: 47308E72.3080307 () dc ! ltu ! se
[Download RAW message or body]

I have looked at hard regs liveness analysis somewhat, and found a few 
things that I haven't got a good idea how to solve yet.  Ideas are 
welcome :-)

To do liveness analysis on hard regs also is not a problem, but:

-  Involving hard regs may cause graphs to be uncolorable (and 
compilation fail) if the
   target writer do not move data to temporaries as he should (should be 
fixed anyway).
-  If optimization is not used, no basic block splitup is done and 
therefore long-range
   liveness analysis will be unusable.  Short-range will work though.

The best way I can think of to handle this is to let the target be 
responsible for all moves
between registers and temporaries and only have very short ranges of 
live hard regs
(only inside basic blocks).  It also turns over the responsibility to 
move arguments
to temporaries to the target instead of in dclargs().

Comments?

-- Ragge

[prev in list] [next in list] [prev in thread] [next in thread]