[Git][ghc/ghc][wip/amg/tweak-co-opt] 2 commits: Tweak isAxiom_maybe to ignore non-newtype axioms
Adam Gundry (@adamgundry)
gitlab at gitlab.haskell.org
Sun Oct 23 20:14:28 UTC 2022
Adam Gundry pushed to branch wip/amg/tweak-co-opt at Glasgow Haskell Compiler / GHC
Commits:
6186ce7c by Adam Gundry at 2022-10-23T21:04:25+01:00
Tweak isAxiom_maybe to ignore non-newtype axioms
See Note [Push transitivity inside newtype axioms only] for an explanation
of the change here. This change substantially improves the performance of
coercion optimization for programs involving transitive type family reductions.
-------------------------
Metric Decrease:
CoOpt_Singletons
LargeRecord
T12227
T12545
T13386
T15703
T5030
T8095
-------------------------
- - - - -
32a2258f by Adam Gundry at 2022-10-23T21:09:39+01:00
Move checkAxInstCo to GHC.Core.Lint
A consequence of the previous change is that checkAxInstCo is no longer
called during coercion optimization, so it can be moved back where it belongs.
- - - - -
2 changed files:
- compiler/GHC/Core/Coercion/Opt.hs
- compiler/GHC/Core/Lint.hs
Changes:
=====================================
compiler/GHC/Core/Coercion/Opt.hs
=====================================
@@ -4,7 +4,6 @@
module GHC.Core.Coercion.Opt
( optCoercion
- , checkAxInstCo
, OptCoercionOpts (..)
)
where
@@ -781,13 +780,12 @@ opt_trans_rule is co1 co2
-- Push transitivity inside axioms
opt_trans_rule is co1 co2
- -- See Note [Why call checkAxInstCo during optimisation]
+ -- See Note [Push transitivity inside axioms]
-- TrPushSymAxR
| Just (sym, con, ind, cos1) <- co1_is_axiom_maybe
, True <- sym
, Just cos2 <- matchAxiom sym con ind co2
, let newAxInst = AxiomInstCo con ind (opt_transList is (map mkSymCo cos2) cos1)
- , Nothing <- checkAxInstCo newAxInst
= fireTransRule "TrPushSymAxR" co1 co2 $ SymCo newAxInst
-- TrPushAxR
@@ -795,7 +793,6 @@ opt_trans_rule is co1 co2
, False <- sym
, Just cos2 <- matchAxiom sym con ind co2
, let newAxInst = AxiomInstCo con ind (opt_transList is cos1 cos2)
- , Nothing <- checkAxInstCo newAxInst
= fireTransRule "TrPushAxR" co1 co2 newAxInst
-- TrPushSymAxL
@@ -803,7 +800,6 @@ opt_trans_rule is co1 co2
, True <- sym
, Just cos1 <- matchAxiom (not sym) con ind co1
, let newAxInst = AxiomInstCo con ind (opt_transList is cos2 (map mkSymCo cos1))
- , Nothing <- checkAxInstCo newAxInst
= fireTransRule "TrPushSymAxL" co1 co2 $ SymCo newAxInst
-- TrPushAxL
@@ -811,7 +807,6 @@ opt_trans_rule is co1 co2
, False <- sym
, Just cos1 <- matchAxiom (not sym) con ind co1
, let newAxInst = AxiomInstCo con ind (opt_transList is cos1 cos2)
- , Nothing <- checkAxInstCo newAxInst
= fireTransRule "TrPushAxL" co1 co2 newAxInst
-- TrPushAxSym/TrPushSymAx
@@ -892,30 +887,87 @@ fireTransRule _rule _co1 _co2 res
= Just res
{-
-Note [Conflict checking with AxiomInstCo]
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-Consider the following type family and axiom:
+Note [Push transitivity inside axioms]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+opt_trans_rule tries to push transitivity inside axioms to deal with cases like
+the following:
+
+ newtype N a = MkN a
+
+ axN :: N a ~R# a
+
+ covar :: a ~R# b
+ co1 = axN <a> :: N a ~R# a
+ co2 = axN <b> :: N b ~R# b
+
+ co :: a ~R# b
+ co = sym co1 ; N covar ; co2
+
+When we are optimising co, we want to notice that the two axiom instantiations
+cancel out. This is implemented by rules such as TrPushSymAxR, which transforms
+ sym (axN <a>) ; N covar
+into
+ sym (axN covar)
+so that TrPushSymAx can subsequently transform
+ sym (axN covar) ; axN <b>
+into
+ covar
+which is much more compact. In some perf test cases this kind of pattern can be
+generated repeatedly during simplification, so it is very important we squash it
+to stop coercions growing exponentially. For more details see the paper:
+
+ Evidence normalisation in System FC
+ Dimitrios Vytiniotis and Simon Peyton Jones
+ RTA'13, 2013
+ https://www.microsoft.com/en-us/research/publication/evidence-normalization-system-fc-2/
+
+
+Note [Push transitivity inside newtype axioms only]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+The optimization described in Note [Push transitivity inside axioms] is possible
+for both newtype and type family axioms. However, for type family axioms it is
+relatively common to have transitive sequences of axioms instantiations, for
+example:
+
+ data Nat = Zero | Suc Nat
+
+ type family Index (n :: Nat) (xs :: [Type]) :: Type where
+ Index Zero (x : xs) = x
+ Index (Suc n) (x : xs) = Index n xs
+
+ axIndex :: { forall x::Type. forall xs::[Type]. Index Zero (x : xs) ~ x
+ ; forall n::Nat. forall x::Type. forall xs::[Type]. Index (Suc n) (x : xs) ~ Index n xs }
+
+ co :: Index (Suc (Suc Zero)) [a, b, c] ~ c
+ co = axIndex[1] <Suc Zero> <a> <[b, c]>
+ ; axIndex[1] <Zero> <b> <[c]>
+ ; axIndex[0] <c> <[]>
+
+Not only are there no cancellation opportunities here, but calling matchAxiom
+repeatedly down the transitive chain is very expensive. Hence we do not attempt
+to push transitivity inside type family axioms. See #8095, !9210 and related tickets.
+
+This is implemented by isAxiom_maybe checking that the axiom is for a newtype
+constructor (i.e. not a type family). Adding this single guard substantially
+improved performance (reduced bytes allocated by more than 10%) for the tests
+CoOpt_Singletons, LargeRecord, T12227, T12545, T13386, T15703, T5030, T8095.
+
+A side benefit is that we do not encounter difficulties with confict checking
+for branched axioms; see Note [Why call checkAxInstCo during optimisation].
+
+There may exist programs that previously relied on pushing transitivity inside
+type family axioms to avoid creating huge coercions, which will regress in
+compile time performance as a result of this change. We do not currently know
+of any examples, but if any come to light we may need to reconsider this
+behaviour.
-type family Equal (a :: k) (b :: k) :: Bool
-type instance where
- Equal a a = True
- Equal a b = False
---
-Equal :: forall k::*. k -> k -> Bool
-axEqual :: { forall k::*. forall a::k. Equal k a a ~ True
- ; forall k::*. forall a::k. forall b::k. Equal k a b ~ False }
-
-We wish to disallow (axEqual[1] <*> <Int> <Int). (Recall that the index is
-0-based, so this is the second branch of the axiom.) The problem is that, on
-the surface, it seems that (axEqual[1] <*> <Int> <Int>) :: (Equal * Int Int ~
-False) and that all is OK. But, all is not OK: we want to use the first branch
-of the axiom in this case, not the second. The problem is that the parameters
-of the first branch can unify with the supplied coercions, thus meaning that
-the first branch should be taken. See also Note [Apartness] in
-"GHC.Core.FamInstEnv".
Note [Why call checkAxInstCo during optimisation]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+NB: The following is no longer relevant, because we no longer push transitivity
+into type family axioms (Note [Push transitivity inside newtype axioms only]).
+It is retained for reference in case we change this behaviour in the future.
+
It is possible that otherwise-good-looking optimisations meet with disaster
in the presence of axioms with multiple equations. Consider
@@ -1006,39 +1058,6 @@ The problem described here was first found in dependent/should_compile/dynamic-p
-}
--- | Check to make sure that an AxInstCo is internally consistent.
--- Returns the conflicting branch, if it exists
--- See Note [Conflict checking with AxiomInstCo]
-checkAxInstCo :: Coercion -> Maybe CoAxBranch
--- defined here to avoid dependencies in GHC.Core.Coercion
--- If you edit this function, you may need to update the GHC formalism
--- See Note [GHC Formalism] in GHC.Core.Lint
-checkAxInstCo (AxiomInstCo ax ind cos)
- = let branch = coAxiomNthBranch ax ind
- tvs = coAxBranchTyVars branch
- cvs = coAxBranchCoVars branch
- incomps = coAxBranchIncomps branch
- (tys, cotys) = splitAtList tvs (map coercionLKind cos)
- co_args = map stripCoercionTy cotys
- subst = zipTvSubst tvs tys `composeTCvSubst`
- zipCvSubst cvs co_args
- target = Type.substTys subst (coAxBranchLHS branch)
- in_scope = mkInScopeSet $
- unionVarSets (map (tyCoVarsOfTypes . coAxBranchLHS) incomps)
- flattened_target = flattenTys in_scope target in
- check_no_conflict flattened_target incomps
- where
- check_no_conflict :: [Type] -> [CoAxBranch] -> Maybe CoAxBranch
- check_no_conflict _ [] = Nothing
- check_no_conflict flat (b at CoAxBranch { cab_lhs = lhs_incomp } : rest)
- -- See Note [Apartness] in GHC.Core.FamInstEnv
- | SurelyApart <- tcUnifyTysFG alwaysBindFun flat lhs_incomp
- = check_no_conflict flat rest
- | otherwise
- = Just b
-checkAxInstCo _ = Nothing
-
-
-----------
wrapSym :: SymFlag -> Coercion -> Coercion
wrapSym sym co | sym = mkSymCo co
@@ -1065,6 +1084,7 @@ isAxiom_maybe (SymCo co)
| Just (sym, con, ind, cos) <- isAxiom_maybe co
= Just (not sym, con, ind, cos)
isAxiom_maybe (AxiomInstCo con ind cos)
+ | isNewTyCon (coAxiomTyCon con) -- See Note [Push transitivity inside newtype axioms only]
= Just (False, con, ind, cos)
isAxiom_maybe _ = Nothing
=====================================
compiler/GHC/Core/Lint.hs
=====================================
@@ -55,7 +55,6 @@ import GHC.Core.TyCo.Ppr ( pprTyVar, pprTyVars )
import GHC.Core.TyCon as TyCon
import GHC.Core.Coercion.Axiom
import GHC.Core.Unify
-import GHC.Core.Coercion.Opt ( checkAxInstCo )
import GHC.Core.Opt.Arity ( typeArity, exprIsDeadEnd )
import GHC.Core.Opt.Monad
@@ -2496,6 +2495,64 @@ lintCoercion (HoleCo h)
= do { addErrL $ text "Unfilled coercion hole:" <+> ppr h
; lintCoercion (CoVarCo (coHoleCoVar h)) }
+
+{-
+Note [Conflict checking with AxiomInstCo]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+Consider the following type family and axiom:
+
+type family Equal (a :: k) (b :: k) :: Bool
+type instance where
+ Equal a a = True
+ Equal a b = False
+--
+Equal :: forall k::*. k -> k -> Bool
+axEqual :: { forall k::*. forall a::k. Equal k a a ~ True
+ ; forall k::*. forall a::k. forall b::k. Equal k a b ~ False }
+
+We wish to disallow (axEqual[1] <*> <Int> <Int). (Recall that the index is
+0-based, so this is the second branch of the axiom.) The problem is that, on
+the surface, it seems that (axEqual[1] <*> <Int> <Int>) :: (Equal * Int Int ~
+False) and that all is OK. But, all is not OK: we want to use the first branch
+of the axiom in this case, not the second. The problem is that the parameters
+of the first branch can unify with the supplied coercions, thus meaning that
+the first branch should be taken. See also Note [Apartness] in
+"GHC.Core.FamInstEnv".
+-}
+
+-- | Check to make sure that an AxInstCo is internally consistent.
+-- Returns the conflicting branch, if it exists
+-- See Note [Conflict checking with AxiomInstCo]
+checkAxInstCo :: Coercion -> Maybe CoAxBranch
+-- defined here to avoid dependencies in GHC.Core.Coercion
+-- If you edit this function, you may need to update the GHC formalism
+-- See Note [GHC Formalism] in GHC.Core.Lint
+checkAxInstCo (AxiomInstCo ax ind cos)
+ = let branch = coAxiomNthBranch ax ind
+ tvs = coAxBranchTyVars branch
+ cvs = coAxBranchCoVars branch
+ incomps = coAxBranchIncomps branch
+ (tys, cotys) = splitAtList tvs (map coercionLKind cos)
+ co_args = map stripCoercionTy cotys
+ subst = zipTvSubst tvs tys `composeTCvSubst`
+ zipCvSubst cvs co_args
+ target = Type.substTys subst (coAxBranchLHS branch)
+ in_scope = mkInScopeSet $
+ unionVarSets (map (tyCoVarsOfTypes . coAxBranchLHS) incomps)
+ flattened_target = flattenTys in_scope target in
+ check_no_conflict flattened_target incomps
+ where
+ check_no_conflict :: [Type] -> [CoAxBranch] -> Maybe CoAxBranch
+ check_no_conflict _ [] = Nothing
+ check_no_conflict flat (b at CoAxBranch { cab_lhs = lhs_incomp } : rest)
+ -- See Note [Apartness] in GHC.Core.FamInstEnv
+ | SurelyApart <- tcUnifyTysFG alwaysBindFun flat lhs_incomp
+ = check_no_conflict flat rest
+ | otherwise
+ = Just b
+checkAxInstCo _ = Nothing
+
+
{-
************************************************************************
* *
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/be10db069b1482f5b24a6add28dbad96abfabf9a...32a2258f2d2652b96a8923b3b7b8ec5748332338
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/be10db069b1482f5b24a6add28dbad96abfabf9a...32a2258f2d2652b96a8923b3b7b8ec5748332338
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/20221023/23af3316/attachment-0001.html>
More information about the ghc-commits
mailing list