[Git][ghc/ghc][wip/romes/no-simple-joinpoints] uncondInline join points for trivial applications without free vars

[Rodrigo Mesquita (@alt-romes)]

Rodrigo Mesquita pushed to branch wip/romes/no-simple-joinpoints at Glasgow Haskell Compiler / GHC


Commits:
980225e3 by Rodrigo Mesquita at 2025-02-05T11:14:20+00:00
uncondInline join points for trivial applications without free vars

- - - - -


1 changed file:

- compiler/GHC/Core/Unfold.hs


Changes:

=====================================
compiler/GHC/Core/Unfold.hs
=====================================
@@ -434,7 +434,7 @@ uncondInline is_join rhs bndrs arity body size
 
 uncondInlineJoin :: [Var] -> CoreExpr -> Bool
 -- See Note [Duplicating join points] point (DJ3) in GHC.Core.Opt.Simplify.Iteration
-uncondInlineJoin _bndrs body
+uncondInlineJoin bndrs body
   | exprIsTrivial body
   = True   -- Nullary constructors, literals
 
@@ -443,9 +443,38 @@ uncondInlineJoin _bndrs body
   , isJoinId v   -- Indirection to another join point; always inline
   = True
 
+  | trivialAppWithoutFreeArgs
+  -- See Note [...]
+  = True
+
   | otherwise
   = False
 
+  where
+    -- Arguments cannot be free Vars!! See Note ...
+    trivialAppWithoutFreeArgs = go body
+    go (App f a)
+
+      -- Go over types
+      | Type _ <- a = go f
+
+      -- 1. If arg is a Var, it must be bound by bndrs
+      | Var v <- a
+      , v `elem` bndrs {- not free -}
+      = go f
+
+      -- 2. If arg is otherwise trivial, continue
+      | exprIsTrivial a = go f
+
+    -- Go over casts (and ticks?)
+    go (Cast e _) = go e
+    go (Tick _ e) = go e
+
+    go (Var _)    = True -- reached base case (function being applied, may be free var)
+
+    go e          = False -- any other case is not trivial app without free args
+
+
 
 sizeExpr :: UnfoldingOpts
          -> Int             -- Bomb out if it gets bigger than this



View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/980225e3dae2f5949ce53fb915b2bd88570309c3

-- 
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/980225e3dae2f5949ce53fb915b2bd88570309c3
You're receiving this email because of your account on gitlab.haskell.org.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-commits/attachments/20250205/a2897ed3/attachment-0001.html>