[GHC] #4101: Primitive constant unfolding

[GHC]

#4101: Primitive constant unfolding
--------------------------------------------+---------------------------
        Reporter:  malosh                   |            Owner:  schyler
            Type:  bug                      |           Status:  new
        Priority:  low                      |        Milestone:  7.6.2
       Component:  Compiler                 |          Version:  6.12.2
      Resolution:                           |         Keywords:
Operating System:  Unknown/Multiple         |     Architecture:  x86
 Type of failure:  Runtime performance bug  |       Difficulty:  Unknown
       Test Case:                           |       Blocked By:
        Blocking:                           |  Related Tickets:
--------------------------------------------+---------------------------

Comment (by carter):

 @schyler

 please please be very careful about anything regarding evaluating floating
 point, I think before we add optimizations and constant propagation for
 floating point computations we really need to enrich GHC's internal data
 model of floating point. a Naive constant progagation alg will not be
 correct for floating point.  I'd in fact argue that any constant
 propagtion stuff first needs that internal model cleaned up first to be
 correct.

 Relatedly: many floating point libs do a lot of work that exploits the
 specific structure of floating point, for example edward kmett's
 compensated lib http://hackage.haskell.org/package/compensated

 I think the first step would be adding better support for writing double#
 and float# literals, perhaps with hexadecimal notation support, as this
 ticket suggests, as a prereq to the subsequent parts of the ticket.
 Perhaps that literal support should be spun out into its own ticket and
 your first task is adding that literals support?

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/4101#comment:13>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler