[GHC] #9661: Branchless ==# is compiled to branchy code

[GHC]

#9661: Branchless ==# is compiled to branchy code
-------------------------------------+-------------------------------------
              Reporter:  dfeuer      |            Owner:
                  Type:  bug         |           Status:  new
              Priority:  normal      |        Milestone:
             Component:  Compiler    |          Version:  7.9
            Resolution:              |         Keywords:
      Operating System:              |     Architecture:  Unknown/Multiple
  Unknown/Multiple                   |       Difficulty:  Unknown
       Type of failure:  Runtime     |       Blocked By:
  performance bug                    |  Related Tickets:
             Test Case:              |
              Blocking:              |
Differential Revisions:              |
-------------------------------------+-------------------------------------

Comment (by jstolarek):

 After some thinking I don't think it's constant folding but more generally
 the special rules for `==#`. As promised here are some hints how to get
 started on this one:

 1. Try to reproduce the problem with a simpler example, eg. using only two
 comparison primops and `orI#`. Try out different combinations of
 comparison primops. See also whether this happens for `andI#`. We want to
 be sure that this only depends on presence of `==#` and not other
 operators (don't forget `/=#`).

 2. Once you have a smaller test case follow the transformation that GHC is
 performing to arrive at the result. Is it turning the simpler test case
 into the final result in one step or are there any intermediate steps?

 3. My guess would be that the culprit is `litEq` function in
 `PrelRules.lhs` module. The comment above it claims to do exactly what
 you've reported here. However, I'm not sure what the fix should be.
 Disabling this is probably not a good idea, since in some cases we really
 want that transformation to happen. We'd need to think more but first
 let's confirm whether `litEq` is the cause or not. You can verify that by
 disabling that rule and seeing whether the code compiles without branches.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9661#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler