[GHC] #7014: RULES for bitwise logic and shift primops

[GHC]

#7014: RULES for bitwise logic and shift primops
---------------------------------+------------------------------------------
    Reporter:  akio              |       Owner:  simonpj         
        Type:  feature request   |      Status:  patch           
    Priority:  normal            |   Milestone:  7.6.1           
   Component:  Compiler          |     Version:  7.4.2           
    Keywords:                    |          Os:  Unknown/Multiple
Architecture:  Unknown/Multiple  |     Failure:  None/Unknown    
  Difficulty:  Unknown           |    Testcase:                  
   Blockedby:                    |    Blocking:                  
     Related:                    |  
---------------------------------+------------------------------------------

Comment(by simonpj):

 Paolo, I'm more or less happy with the patch, with the following
 suggestions:

  * The local function `primop_rule` now always returns a singleton list;
 any multi-rule stuff is done within that single rule.  So I suggest making
 the top-level `primOpRule` function do the pattern matching, and return
 one rule:
 {{{
 primOpRule :: PrimOp -> Name -> CoreRule
 primOpRule IntAddOp nm = mkPrimOpRule nm 2 [ binaryLit (intOp2 (+))
                                            , identity zeroi ]
 ..etc...
 }}}

  * That turns `relop` and `rules` into top-level functions, which can have
 type signatures and helpful names.

  * I'd prefer to eta-expand `rules`.  Otherwise the `msum` is a bit
 cryptic.

  * You may be able to use `convFloating` less often, by having a variant
 of `getLiteral` or whatever for floating, et `getFloatingLiteral`, used
 for floating primops.

  * Can you comment the `Int` argument of `getLiteral`?

 Then go ahead and commit. Thanks!

 Simon

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/7014#comment:10>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler