diff options
| author | Simon Peyton Jones <simonpj@microsoft.com> | 2019-11-08 17:49:35 +0000 |
|---|---|---|
| committer | Marge Bot <ben+marge-bot@smart-cactus.org> | 2020-04-02 01:45:58 -0400 |
| commit | b943b25d0786da64031ac63ddf9b4574182057bb (patch) | |
| tree | 529ac144237d1c4d7524360374fa8b33a6bd5ee4 /compiler/GHC/CoreToStg | |
| parent | 3c09f636a459f50119bfbb5bf798b9a9e19eb464 (diff) | |
| download | haskell-b943b25d0786da64031ac63ddf9b4574182057bb.tar.gz | |
Re-engineer the binder-swap transformation
The binder-swap transformation is implemented by the occurrence
analyser -- see Note [Binder swap] in OccurAnal. However it had
a very nasty corner in it, for the case where the case scrutinee
was a GlobalId. This led to trouble and hacks, and ultimately
to #16296.
This patch re-engineers how the occurrence analyser implements
the binder-swap, by actually carrying out a substitution rather
than by adding a let-binding. It's all described in
Note [The binder-swap substitution].
I did a few other things along the way
* Fix a bug in StgCse, which could allow a loop breaker to be CSE'd
away. See Note [Care with loop breakers] in StgCse. I think it can
only show up if occurrence analyser sets up bad loop breakers, but
still.
* Better commenting in SimplUtils.prepareAlts
* A little refactoring in CoreUnfold; nothing significant
e.g. rename CoreUnfold.mkTopUnfolding to mkFinalUnfolding
* Renamed CoreSyn.isFragileUnfolding to hasCoreUnfolding
* Move mkRuleInfo to CoreFVs
We observed respectively 4.6% and 5.9% allocation decreases for the following
tests:
Metric Decrease:
T9961
haddock.base
Diffstat (limited to 'compiler/GHC/CoreToStg')
| -rw-r--r-- | compiler/GHC/CoreToStg/Prep.hs | 14 |
1 files changed, 5 insertions, 9 deletions
diff --git a/compiler/GHC/CoreToStg/Prep.hs b/compiler/GHC/CoreToStg/Prep.hs index b6a14b4af5..5cdf084a33 100644 --- a/compiler/GHC/CoreToStg/Prep.hs +++ b/compiler/GHC/CoreToStg/Prep.hs @@ -347,10 +347,7 @@ The way we fix this is to: * In cloneBndr, drop all unfoldings/rules * In deFloatTop, run a simple dead code analyser on each top-level - RHS to drop the dead local bindings. For that call to OccAnal, we - disable the binder swap, else the occurrence analyser sometimes - introduces new let bindings for cased binders, which lead to the bug - in #5433. + RHS to drop the dead local bindings. The reason we don't just OccAnal the whole output of CorePrep is that the tidier ensures that all top-level binders are GlobalIds, so they @@ -1316,14 +1313,13 @@ deFloatTop :: Floats -> [CoreBind] deFloatTop (Floats _ floats) = foldrOL get [] floats where - get (FloatLet b) bs = occurAnalyseRHSs b : bs - get (FloatCase body var _ _ _) bs - = occurAnalyseRHSs (NonRec var body) : bs + get (FloatLet b) bs = get_bind b : bs + get (FloatCase body var _ _ _) bs = get_bind (NonRec var body) : bs get b _ = pprPanic "corePrepPgm" (ppr b) -- See Note [Dead code in CorePrep] - occurAnalyseRHSs (NonRec x e) = NonRec x (occurAnalyseExpr_NoBinderSwap e) - occurAnalyseRHSs (Rec xes) = Rec [(x, occurAnalyseExpr_NoBinderSwap e) | (x, e) <- xes] + get_bind (NonRec x e) = NonRec x (occurAnalyseExpr e) + get_bind (Rec xes) = Rec [(x, occurAnalyseExpr e) | (x, e) <- xes] --------------------------------------------------------------------------- |