#3697: Method selectors aren't floated out of loops
---------------------------------+------------------------------------------
Reporter: rl | Owner:
Type: bug | Status: new
Priority: normal | Component: Compiler
Version: 6.13 | Keywords:
Os: Unknown/Multiple | Testcase:
Architecture: Unknown/Multiple | Failure: Runtime performance bug
---------------------------------+------------------------------------------
Here is a small example:
{{{
foo :: Num a => [a] -> [a]
foo as = filter (/=0) (map (\x -> x-x) as)
}}}
Here is the code that the current HEAD generates for the loop:
{{{
go_smy =
\ (ds_akN :: [a_aiw]) ->
case ds_akN of _ {
[] -> GHC.Types.[] @ a_aiw;
: y_akS ys_akT ->
let {
x_smA :: a_aiw
x_smA = GHC.Num.- @ a_aiw $dNum_ajk y_akS y_akS } in
case GHC.Classes./= @ a_aiw lvl_smu x_smA lit_smw of _ {
GHC.Bool.False -> go_smy ys_akT;
GHC.Bool.True -> GHC.Types.: @ a_aiw x_smA (go_smy ys_akT)
}
}}}
Note that the `Eq` dictionary is inspected by `(/=)` in every loop
iteration. Floating out `GHC.Classes./= @ a_aiw lvl_smu` (and, perhaps,
`GHC.Num.- @ a_aiw $dNum_ajk`) should solve this. In fact, I wonder why
that isn't happening already. Do method selectors have a wrong arity?
FWIW, in GHC 6.10 `LiberateCase` moves the method selection out of the
loop (not ideal, but it gets the job done) so !NoSlow shows this as a
slight performance regression.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/3697>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
