diff options
| author | Sebastian Graf <sebastian.graf@kit.edu> | 2019-02-19 13:52:11 +0100 |
|---|---|---|
| committer | Marge Bot <ben+marge-bot@smart-cactus.org> | 2019-03-07 20:44:08 -0500 |
| commit | 1675d40afe07b9c414eaa37d85819f37f8420118 (patch) | |
| tree | 0e0fe32fdf1a70a90e2c531a89b0a16b07fbad20 /utils/genprimopcode | |
| parent | 068b7e983f4a0b35f453aa5e609998efd0c3f334 (diff) | |
| download | haskell-1675d40afe07b9c414eaa37d85819f37f8420118.tar.gz | |
Always do the worker/wrapper split for NOINLINEs
Trac #10069 revealed that small NOINLINE functions didn't get split
into worker and wrapper. This was due to `certainlyWillInline`
saying that any unfoldings with a guidance of `UnfWhen` inline
unconditionally. That isn't the case for NOINLINE functions, so we
catch this case earlier now.
Nofib results:
--------------------------------------------------------------------------------
Program Allocs Instrs
--------------------------------------------------------------------------------
fannkuch-redux -0.3% 0.0%
gg +0.0% +0.1%
maillist -0.2% -0.2%
minimax 0.0% -0.8%
--------------------------------------------------------------------------------
Min -0.3% -0.8%
Max +0.0% +0.1%
Geometric Mean -0.0% -0.0%
Fixes #10069.
-------------------------
Metric Increase:
T9233
-------------------------
Diffstat (limited to 'utils/genprimopcode')
| -rw-r--r-- | utils/genprimopcode/Main.hs | 8 |
1 files changed, 7 insertions, 1 deletions
diff --git a/utils/genprimopcode/Main.hs b/utils/genprimopcode/Main.hs index d7ae9ffe01..3427a1ebae 100644 --- a/utils/genprimopcode/Main.hs +++ b/utils/genprimopcode/Main.hs @@ -611,7 +611,13 @@ gen_wrappers (Info _ entries) = "{-# LANGUAGE MagicHash, NoImplicitPrelude, UnboxedTuples #-}\n" -- Dependencies on Prelude must be explicit in libraries/base, but we -- don't need the Prelude here so we add NoImplicitPrelude. - ++ "{-# OPTIONS_GHC -Wno-deprecations #-}\n" + ++ "{-# OPTIONS_GHC -Wno-deprecations -O0 #-}\n" + -- No point in optimising this at all. + -- Performing WW on this module is harmful even, two reasons: + -- 1. Inferred strictness signatures are all bottom, which is a lie + -- 2. Doing the worker/wrapper split based on that information will + -- introduce references to Control.Exception.Base.absentError, + -- which isn't available at this point. ++ "module GHC.PrimopWrappers where\n" ++ "import qualified GHC.Prim\n" ++ "import GHC.Tuple ()\n" |