GitLab

Simon Peyton Jones pushed to branch wip/T26548 at Glasgow Haskell Compiler / GHC

Commits:

6df6e234

by Simon Peyton Jones at 2025-11-27T11:33:06+00:00

Revert making Reduction stricter

It makes things slightly worse I think.

7b21eaef

by Simon Peyton Jones at 2025-11-27T11:33:35+00:00

Make inlining a tiny bit more eager

In particular
     x = mkSymMCo mco
where mco is evaluated.  We want that to inline, especially if the
let is strict.  Makes a significant difference in Rewrite.hs,
Test case T9872b

3 changed files:

compiler/GHC/Core/Opt/Simplify/Inline.hs
compiler/GHC/Core/Opt/Simplify/Utils.hs
compiler/GHC/Core/Reduction.hs

Changes:

compiler/GHC/Core/Opt/Simplify/Inline.hs

@@ -390,7 +390,7 @@ tryUnfolding env logger id lone_variable arg_infos
             -- be *something* interesting about some argument, or the
             -- result context, to make it worth inlining
      calc_some_benefit :: Arity -> Bool -> Bool   -- The Arity is the number of args
 -                                         -- expected by the unfolding
 +                                                 -- expected by the unfolding
      calc_some_benefit uf_arity is_inline
         | not saturated = interesting_args       -- Under-saturated
                                          -- Note [Unsaturated applications]

compiler/GHC/Core/Opt/Simplify/Utils.hs

@@ -910,7 +910,7 @@ interestingCallContext env cont
          -- in GHC.Core.Unfold
      interesting (StrictArg { sc_fun = fun }) = strictArgContext fun
 -    interesting (StrictBind {})              = BoringCtxt
 +    interesting (StrictBind {})              = RhsCtxt NonRecursive
      interesting (Stop _ cci _)               = cci
      interesting (TickIt _ k)                 = interesting k
      interesting (ApplyToTy { sc_cont = k })  = interesting k
@@ -1032,7 +1032,7 @@ interestingArg env e = go env 0 e
         | n > 0         = NonTrivArg -- Saturated or unknown call
         | otherwise  -- n==0, no value arguments; look for an interesting unfolding
         = case idUnfolding v of
 -           OtherCon [] -> TrivArg      -- It's evaluated, but that's all we know
 +--           OtherCon [] -> NonTrivArg   -- It's evaluated, but that's all we know
             OtherCon _  -> NonTrivArg   -- Evaluated and we know it isn't these constructors
                -- See Note [OtherCon and interestingArg]
             DFunUnfolding {} -> ValueArg   -- We konw that idArity=0

compiler/GHC/Core/Reduction.hs

@@ -94,7 +94,7 @@ but in fact `reductionOriginalType` is very seldom used, so it's not worth it.
  -- See Note [The Reduction type].
  data Reduction =
    Reduction
 -    { reductionCoercion    :: !Coercion
 +    { reductionCoercion    :: Coercion
      , reductionReducedType :: !Type
+     }
  -- N.B. the 'Coercion' field must be lazy: see for instance GHC.Tc.Solver.Rewrite.rewrite_tyvar2