#12990: Partially applied constructors with unpacked fields simplified badly -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1 Resolution: | Keywords: Inlining 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 dfeuer): Replying to [comment:3 simonpj]:
In `MkId` line 511 we have {{{ wrap_unf = mkInlineUnfolding (Just wrap_arity) wrap_rhs }}} The `Just wrap_arity` is what prevents GHC inlining the wrapper until it is saturated. Change it to `Nothing` to allow a partially-applied wrapper to inline.
Worth a try. I doubt it'd have any bad effects.
There's a comment right above that I don't know enough about GHC to interpret:
The Cpr info can be important inside INLINE rhss, where the wrapper constructor isn't inlined. And the argument strictness can be important too; we may not inline a constructor when it is partially applied. For example: {{{#!hs data W = C !Int !Int !Int ...(let w = C x in ...(w p q)...)... }}} we want to see that w is strict in its two arguments
Does this point to a reason not to inline partially-applied wrappers? If not, does it point to some additional change that should be made in parallel? Or does it become obsolete with this change? -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/12990#comment:6 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler