Haifa buglet

[Jeffrey A Law]

  In message < 200001121852.SAA09843@cam-mail2.cambridge.arm.com >you write:
  > This is a multipart MIME message.
  > 
  > --==_Exmh_13719096420
  > Content-Type: text/plain; charset=us-ascii
  > 
  > 
  > Jeff's recent change in the call extract_bitlst changed the second 
  > argument to be es (presumably on the assumption that this had been 
  > recently initialized to edgeset_size).  Unfortunately, this is wrong, 
  > since es is then used as a loop iterator so by the time of the call is 
  > always -1.  Patch below installed also adds a declartion for bitlen.
  > 
  > This fixes a bootstrap error on the ARM when the scheduler incorrectly 
  > steals an increment insn from another block even though the edge is 
  > conditional and the reg is live across both exit routes.
Thanks.  I bet this also fixes the bootstrap error I was just tracking down
on the PA.

I've got a nice small testcase for this one :-)

jeff