Simon Jakobi pushed to branch wip/sjakobi/udfm-placement at Glasgow Haskell Compiler / GHC

Commits:

• 966e260a
  by Simon Jakobi at 2026-07-08T07:00:19+02:00

  UDFM: don't INLINE udfmToList (fixes T17836 alloc regression)
  
  Commit 6990e63f made udfmToList INLINE so its n <= 1 case would fuse with
  its consumer. But udfmToList's one hot consumer -- traverseUSDFM in the
  pattern-match checker -- doesn't fuse, and inlining the size dispatch into
  it regressed the T17836 ghc/alloc perf test.
  
  Unlike eltsUDFM, which is called on ~2M singleton maps by fusing consumers
  in T13719, udfmToList fuses too rarely (~60k singleton calls) to earn its
  INLINE.
  
  Assisted-by: Claude Opus 4.8
  

1 changed file:

• compiler/GHC/Types/Unique/DFM.hs

Changes:

 -- as this already incurs most of the cost of returning the full list.

 -- See Note [Cost of deterministic iteration].

 udfmToList :: UniqDFM key elt -> [(Unique, elt)]

-{-# INLINE udfmToList #-}  -- so the small case is a good producer

+-- NB: no INLINE, unlike eltsUDFM. udfmToList's one hot consumer is

+-- traverseUSDFM in the pattern-match checker, which doesn't fuse. Inlining

+-- the size dispatch into it regresses T17836.

 udfmToList (UDFM m ub)

   -- n <= 1: any order is trivially tag order, so read straight off the map

   | M.compareSize m 1 /= GT =

-      build (\c n ->

-        M.foldrWithKey (\k tv r -> c (mkUniqueGrimily k, taggedFst tv) r) n m)

+      M.foldrWithKey (\k tv r -> (mkUniqueGrimily k, taggedFst tv) : r) [] m

   | otherwise = to_list_nonempty m ub

 -- Precondition: m is non-empty