[GHC] #10069: CPR related performance issue

[GHC]

#10069: CPR related performance issue
-------------------------------------+-------------------------------------
        Reporter:  pacak             |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.6.2
      Resolution:                    |             Keywords:
                                     |  DemandAnalysis
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Runtime           |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by simonpj):

 >  So: If we WW fc, we should also WW c1, otherwise we end up with bad
 code.

 Quite right. Why does that not happen?  It's because of
 {{{
 tryWW dflags fam_envs is_rec fn_id rhs
   -- See Note [Worker-wrapper for NOINLINE functions]

   | Just stable_unf <- certainlyWillInline dflags fn_info
   = return [ (fn_id `setIdUnfolding` stable_unf, rhs) ]
         -- See Note [Don't w/w INLINE things]
         -- See Note [Don't w/w inline small non-loop-breaker things]
 }}}
 The Notes are informative.

 What happens here is that

 * `certainlyWillInline` returns True for `c1`, despite the NOINLINE pragma
 * We set a stable unfolding, for reasons described in `Note [Don't w/w
 inline small non-loop-breaker things]`
 * But alas the NOINLINE stays there and prevents `c1` inlining.

 I think we should always create a wrapper for NOINLINE things, just as if
 they were big.  We already have technology for transferring the NOINLINE
 pragma to the worker; see
 * `Note [Worker-wrapper for INLINABLE functions]`
 * `Note [Worker activation]`, which even talks about NOINLINE
 specifically.

 So the bottom line is, I think, that the `certainlyWillInline` test should
 return False for NOINLINE functions.  This call in `WorkWrap` is its only
 call site, so the change should be easy.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10069#comment:33>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler