exprIsExpandable bug

Consider this program, derived from T5623 (c.f. #5623 (closed))

{-# LANGUAGE MagicHash, UnboxedTuples #-}
{-# OPTIONS_GHC -funfolding-use-threshold=10 #-}

module T5623 where

import GHC.Ptr
import GHC.Prim
import GHC.Exts

foo :: Ptr Float -> State# RealWorld -> (# State# RealWorld, Float #)
-- foo p = liftM2 (+) (peekElemOff q 0) (peekElemOff q 1)
foo p s = case q :: Ptr Float of { Ptr a1 ->
          case readFloatOffAddr# a1 0# s of { (# s1, f1 #) ->
          case q :: Ptr Float of { Ptr a2 ->
          case readFloatOffAddr# a2 1# s of { (# s2, f2 #) ->
          (# s2, F# (plusFloat# f1 f2) #) }}}}
  where
    q :: Ptr a  -- Polymorphic
    q = p `plusPtr` 4

The -funfolding-use-threshold=10 is important because it makes the worker/wrapper pass do a w/w split for foo. (Otherwise it's too small for w/w.)

The optimised core looks like

foo =
  \ (w_s2TO :: Ptr Float) (w1_s2TP :: State# RealWorld) ->
    case w_s2TO of { Ptr ww1_s2TS ->
    case readFloatOffAddr#
           @ RealWorld (plusAddr# ww1_s2TS 4#) 0# w1_s2TP
    of
    { (# ipv_s2ST, ipv1_s2SU #) ->
    case readFloatOffAddr#
           @ RealWorld (plusAddr# ww1_s2TS 4#) 1# w1_s2TP
    of
    { (# ipv2_s2SY, ipv3_s2SZ #) ->
    (# ipv2_s2SY, GHC.Types.F# (plusFloat# ipv1_s2SU ipv3_s2SZ) #)
    }    }    }

Yikes! Look at that duplicated plusAddr#!

It turned out to be a hard-to-trigger bug in exprIsExpandable.

Trac metadata

Trac field	Value
Version	8.0.1
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture