[GHC] #8456: Control flow optimisations duplicate blocks

[GHC]

#8456: Control flow optimisations duplicate blocks
------------------------------+--------------------------------------------
       Reporter:  jstolarek   |             Owner:  jstolarek
           Type:  bug         |            Status:  new
       Priority:  high        |         Milestone:
      Component:  Compiler    |           Version:  7.7
       Keywords:              |  Operating System:  Unknown/Multiple
   Architecture:              |   Type of failure:  Runtime performance bug
  Unknown/Multiple            |         Test Case:
     Difficulty:  Unknown     |          Blocking:  8275
     Blocked By:              |
Related Tickets:              |
------------------------------+--------------------------------------------
 During work on #8275 I observed that control flow optimisation pass in the
 Cmm pipeline duplicates block that does the stack check, which is
 completely redundant. There is at least one bug in !CmmContFlowOpt module.
 Consider this:

 {{{
 L1: goto L2
 L2: whatever
 L3: goto L1
 }}}

 We are processing blocks from the end. When we reach L3 first guard in
 `maybe_concat` function (!CmmContFlowOpt.hs, line 123) will turn that
 blocks into:

 {{{
 L1: goto L2
 L2: whatever
 L3: goto L2
 }}}

 However, the number of predecessors of L2 block is not updated because
 `backEdges` is computed once before we run `maybe_concat` and is not
 updated when we make changes to the block structure.

 Another issue, which I have not yet encountered in practice but I believe
 may arise as well, comes from the fact that we may map one label to a
 different one, but again we don't that take into account when determining
 the number of predecessors.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8456>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler