[Git][ghc/ghc][wip/spj-unf-size] Adjust thresholds

[Simon Peyton Jones (@simonpj)]

Simon Peyton Jones pushed to branch wip/spj-unf-size at Glasgow Haskell Compiler / GHC


Commits:
ddd17134 by Simon Peyton Jones at 2023-11-19T23:32:16+00:00
Adjust thresholds

- - - - -


2 changed files:

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


Changes:

=====================================
compiler/GHC/Core/Opt/Simplify/Inline.hs
=====================================
@@ -660,10 +660,10 @@ exprDigest env e = go env e []
                      -- attract that ScrutOf discount!  We want liftM2 to be inlined
                      -- in its use in the liftA2 method of instance Applicative (ST s)
                      --
-                     -- Actually in specrtal/puzzle I found that we got a big (40%!)
+                     -- Actually in spectral/puzzle I found that we got a big (40%!)
                      -- benefit from    let newDest = ... in case (notSeen newDest) of ...
                      -- We want to inline notSeen.  The argument has structure (its RHS)
-                     -- and in fat if we inline notSeen, newDest turns into a thunk
+                     -- and in fact if we inline notSeen, newDest turns into a thunk
                      -- (SPJ GHC log 13 Nov).
 
       | Just rhs <- expandUnfolding_maybe unfolding


=====================================
compiler/GHC/Core/Unfold.hs
=====================================
@@ -206,7 +206,7 @@ defaultUnfoldingOpts = UnfoldingOpts
       -- inline into Csg.calc (The unfolding for sqr never makes it
       -- into the interface file.)
 
-   , unfoldingUseThreshold   = 80
+   , unfoldingUseThreshold   = 70
       -- Adjusted 90 -> 80 when adding discounts for free variables which
       -- generally make things more likely to inline.  Reducing the threshold
       -- eliminates some undesirable compile-time regressions (e.g. T10412a)
@@ -214,7 +214,7 @@ defaultUnfoldingOpts = UnfoldingOpts
       -- Previously: adjusted upwards in #18282, when I reduced
       -- the result discount for constructors.
 
-   , unfoldingFunAppDiscount = 60
+   , unfoldingFunAppDiscount = 30
       -- Be fairly keen to inline a function if that means
       -- we'll be able to pick the right method from a dictionary
 
@@ -882,12 +882,11 @@ conSize :: DataCon -> Int -> ExprTree
 conSize dc n_val_args
   | isUnboxedTupleDataCon dc
   = etZero     -- See Note [Unboxed tuple size and result discount]
+  | n_val_args == 0    -- Like variables
+  = etZero
   | otherwise  -- See Note [Constructor size and result discount]
-  = ExprTree { et_size = size, et_wc_tot = size
+  = ExprTree { et_size = 10, et_wc_tot = 10
              , et_cases = emptyBag, et_ret = 10 }
-  where
-    size | n_val_args == 0 = 0  -- Like variables
-         | otherwise       = 10
 
 primOpSize :: PrimOp -> Int -> Size
 primOpSize op n_val_args



View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/ddd17134826249bbe87c370dbcabb12fc7a6c5d9

-- 
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/ddd17134826249bbe87c370dbcabb12fc7a6c5d9
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/20231119/f596476a/attachment-0001.html>