[Git][ghc/ghc][wip/T21754] Boxity: Don't update Boxity unless worker/wrapper follows (#21754)
Sebastian Graf (@sgraf812)
gitlab at gitlab.haskell.org
Fri Sep 30 16:42:24 UTC 2022
Sebastian Graf pushed to branch wip/T21754 at Glasgow Haskell Compiler / GHC
Commits:
f5e8f493 by Sebastian Graf at 2022-09-30T18:42:13+02:00
Boxity: Don't update Boxity unless worker/wrapper follows (#21754)
A small refactoring in our Core Opt pipeline and some new functions for
transfering argument boxities from one signature to another to facilitate
`Note [Don't change boxity without worker/wrapper]`.
Fixes #21754.
- - - - -
9 changed files:
- compiler/GHC/Core/Opt/DmdAnal.hs
- compiler/GHC/Core/Opt/Pipeline.hs
- compiler/GHC/Core/Opt/Pipeline/Types.hs
- compiler/GHC/Driver/Config/Core/Lint.hs
- compiler/GHC/Types/Demand.hs
- testsuite/tests/simplCore/should_compile/spec-inline.stderr
- + testsuite/tests/stranal/sigs/T21754.hs
- + testsuite/tests/stranal/sigs/T21754.stderr
- testsuite/tests/stranal/sigs/all.T
Changes:
=====================================
compiler/GHC/Core/Opt/DmdAnal.hs
=====================================
@@ -59,9 +59,18 @@ _ = pprTrace -- Tired of commenting out the import all the time
-- | Options for the demand analysis
data DmdAnalOpts = DmdAnalOpts
- { dmd_strict_dicts :: !Bool -- ^ Use strict dictionaries
- , dmd_unbox_width :: !Int -- ^ Use strict dictionaries
+ { dmd_strict_dicts :: !Bool
+ -- ^ Value of `-fdicts-strict` (on by default).
+ -- When set, all functons are implicitly strict in dictionary args.
+ , dmd_do_boxity :: !Bool
+ -- ^ Governs whether the analysis should update boxity signatures.
+ -- See Note [Don't change boxity without worker/wrapper].
+ , dmd_unbox_width :: !Int
+ -- ^ Value of `-fdmd-unbox-width`.
+ -- See Note [Unboxed demand on function bodies returning small products]
, dmd_max_worker_args :: !Int
+ -- ^ Value of `-fmax-worker-args`.
+ -- Don't unbox anything if we end up with more than this many args.
}
-- This is a strict alternative to (,)
@@ -146,6 +155,40 @@ unforced thunks in demand or strictness information; and it is the
most memory-intensive part of the compilation process, so this added
seqBinds makes a big difference in peak memory usage.
+Note [Don't change boxity without worker/wrapper]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+Consider (T21754)
+ f n = n+1
+ {-# NOINLINE f #-}
+With `-fno-worker-wrapper`, we should not give `f` a boxity signature that says
+that it unboxes its argument! Client modules would never be able to cancel away
+the box for n. Likewise we shouldn't give `f` the CPR property.
+
+Similarly, in the last run of DmdAnal before codegen (which does not have a
+worker/wrapper phase) we should not change boxity in any way. Remember: an
+earlier result of the demand analyser, complete with worker/wrapper, has aleady
+given a demand signature (with boxity info) to the function.
+(The "last run" is mainly there to attach demanded-once info to let-bindings.)
+
+In general, we should not run Note [Boxity analysis] unless worker/wrapper
+follows to exploit the boxity and make sure that calling modules can observe the
+reported boxity.
+
+Hence DmdAnal is configured by a flag `dmd_do_boxity` that is True only
+if worker/wrapper follows after DmdAnal. If it is not set, and the signature
+is not subject to Note [Boxity for bottoming functions], DmdAnal tries
+to transfer over the previous boxity to the new demand signature, in
+`setIdDmdAndBoxSig`.
+
+Why isn't CprAnal configured with a similar flag? Because if we aren't going to
+do worker/wrapper we don't run CPR analysis at all. (see GHC.Core.Opt.Pipeline)
+
+It might be surprising that we only try to preserve *arg* boxity, not boxity on
+FVs. But FV demands won't make it into interface files anyway, so it's a waste
+of energy.
+Besides, W/W zaps the `DmdEnv` portion of a signature, so we don't know the old
+boxity to begin with; see Note [Zapping DmdEnv after Demand Analyzer].
+
Note [Analysing top-level bindings]
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Consider a CoreProgram like
@@ -257,6 +300,16 @@ setBindIdDemandInfo top_lvl id dmd = setIdDemandInfo id $ case top_lvl of
TopLevel | not (isInterestingTopLevelFn id) -> topDmd
_ -> dmd
+-- | Update the demand signature, but be careful not to change boxity info if
+-- `dmd_do_boxity` is True or if the signature is bottom.
+-- See Note [Don't change boxity without worker/wrapper]
+-- and Note [Boxity for bottoming functions].
+setIdDmdAndBoxSig :: DmdAnalOpts -> Id -> DmdSig -> Id
+setIdDmdAndBoxSig opts id sig = setIdDmdSig id $
+ if dmd_do_boxity opts || isBottomingSig sig
+ then sig
+ else transferArgBoxityDmdSig (idDmdSig id) sig
+
-- | Let bindings can be processed in two ways:
-- Down (RHS before body) or Up (body before RHS).
-- This function handles the up variant.
@@ -1018,7 +1071,8 @@ dmdAnalRhsSig top_lvl rec_flag env let_dmd id rhs
sig = mkDmdSigForArity threshold_arity (DmdType sig_fv final_rhs_dmds rhs_div)
- final_id = id `setIdDmdSig` sig
+ opts = ae_opts env
+ final_id = setIdDmdAndBoxSig opts id sig
!final_env = extendAnalEnv top_lvl env final_id sig
-- See Note [Aggregated demand for cardinality]
@@ -1858,8 +1912,9 @@ dmdFix :: TopLevelFlag
dmdFix top_lvl env let_dmd orig_pairs
= loop 1 initial_pairs
where
+ opts = ae_opts env
-- See Note [Initialising strictness]
- initial_pairs | ae_virgin env = [(setIdDmdSig id botSig, rhs) | (id, rhs) <- orig_pairs ]
+ initial_pairs | ae_virgin env = [(setIdDmdAndBoxSig opts id botSig, rhs) | (id, rhs) <- orig_pairs ]
| otherwise = orig_pairs
-- If fixed-point iteration does not yield a result we use this instead
=====================================
compiler/GHC/Core/Opt/Pipeline.hs
=====================================
@@ -150,7 +150,7 @@ getCoreToDo dflags rule_base extra_vars
maybe_rule_check phase = runMaybe rule_check (CoreDoRuleCheck phase)
maybe_strictness_before (Phase phase)
- | phase `elem` strictnessBefore dflags = CoreDoDemand
+ | phase `elem` strictnessBefore dflags = CoreDoDemand False
maybe_strictness_before _
= CoreDoNothing
@@ -171,8 +171,8 @@ getCoreToDo dflags rule_base extra_vars
simpl_gently = CoreDoSimplify $ initSimplifyOpts dflags extra_vars max_iter
(initGentleSimplMode dflags) rule_base
- dmd_cpr_ww = if ww_on then [CoreDoDemand,CoreDoCpr,CoreDoWorkerWrapper]
- else [CoreDoDemand,CoreDoCpr]
+ dmd_cpr_ww = if ww_on then [CoreDoDemand True,CoreDoCpr,CoreDoWorkerWrapper]
+ else [CoreDoDemand False] -- NB: No CPR! See Note [Don't change boxity without worker/wrapper]
demand_analyser = (CoreDoPasses (
@@ -340,7 +340,7 @@ getCoreToDo dflags rule_base extra_vars
-- has run at all. See Note [Final Demand Analyser run] in GHC.Core.Opt.DmdAnal
-- It is EXTREMELY IMPORTANT to run this pass, otherwise execution
-- can become /exponentially/ more expensive. See #11731, #12996.
- runWhen (strictness || late_dmd_anal) CoreDoDemand,
+ runWhen (strictness || late_dmd_anal) (CoreDoDemand False),
maybe_rule_check FinalPhase,
@@ -491,8 +491,8 @@ doCorePass pass guts = do
CoreDoExitify -> {-# SCC "Exitify" #-}
updateBinds exitifyProgram
- CoreDoDemand -> {-# SCC "DmdAnal" #-}
- updateBindsM (liftIO . dmdAnal logger dflags fam_envs (mg_rules guts))
+ CoreDoDemand before_ww -> {-# SCC "DmdAnal" #-}
+ updateBindsM (liftIO . dmdAnal logger before_ww dflags fam_envs (mg_rules guts))
CoreDoCpr -> {-# SCC "CprAnal" #-}
updateBindsM (liftIO . cprAnalProgram logger fam_envs)
@@ -557,10 +557,11 @@ ruleCheckPass current_phase pat guts = do
rule_fn (mg_binds guts))
return guts
-dmdAnal :: Logger -> DynFlags -> FamInstEnvs -> [CoreRule] -> CoreProgram -> IO CoreProgram
-dmdAnal logger dflags fam_envs rules binds = do
+dmdAnal :: Logger -> Bool -> DynFlags -> FamInstEnvs -> [CoreRule] -> CoreProgram -> IO CoreProgram
+dmdAnal logger before_ww dflags fam_envs rules binds = do
let !opts = DmdAnalOpts
{ dmd_strict_dicts = gopt Opt_DictsStrict dflags
+ , dmd_do_boxity = before_ww -- only run Boxity Analysis immediately preceding WW
, dmd_unbox_width = dmdUnboxWidth dflags
, dmd_max_worker_args = maxWorkerArgs dflags
}
=====================================
compiler/GHC/Core/Opt/Pipeline/Types.hs
=====================================
@@ -45,7 +45,8 @@ data CoreToDo -- These are diff core-to-core passes,
| CoreDoStaticArgs
| CoreDoCallArity
| CoreDoExitify
- | CoreDoDemand
+ | CoreDoDemand Bool -- Bool: Do worker/wrapper afterwards?
+ -- See Note [Don't change boxity without worker/wrapper]
| CoreDoCpr
| CoreDoWorkerWrapper
| CoreDoSpecialising
@@ -74,7 +75,8 @@ instance Outputable CoreToDo where
ppr CoreDoStaticArgs = text "Static argument"
ppr CoreDoCallArity = text "Called arity analysis"
ppr CoreDoExitify = text "Exitification transformation"
- ppr CoreDoDemand = text "Demand analysis"
+ ppr (CoreDoDemand True) = text "Demand analysis (including Boxity)"
+ ppr (CoreDoDemand False) = text "Demand analysis"
ppr CoreDoCpr = text "Constructed Product Result analysis"
ppr CoreDoWorkerWrapper = text "Worker Wrapper binds"
ppr CoreDoSpecialising = text "Specialise"
=====================================
compiler/GHC/Driver/Config/Core/Lint.hs
=====================================
@@ -83,7 +83,7 @@ coreDumpFlag CoreLiberateCase = Just Opt_D_verbose_core2core
coreDumpFlag CoreDoStaticArgs = Just Opt_D_verbose_core2core
coreDumpFlag CoreDoCallArity = Just Opt_D_dump_call_arity
coreDumpFlag CoreDoExitify = Just Opt_D_dump_exitify
-coreDumpFlag CoreDoDemand = Just Opt_D_dump_stranal
+coreDumpFlag (CoreDoDemand {}) = Just Opt_D_dump_stranal
coreDumpFlag CoreDoCpr = Just Opt_D_dump_cpranal
coreDumpFlag CoreDoWorkerWrapper = Just Opt_D_dump_worker_wrapper
coreDumpFlag CoreDoSpecialising = Just Opt_D_dump_spec
=====================================
compiler/GHC/Types/Demand.hs
=====================================
@@ -64,7 +64,8 @@ module GHC.Types.Demand (
-- * Demand signatures
DmdSig(..), mkDmdSigForArity, mkClosedDmdSig, mkVanillaDmdSig,
splitDmdSig, dmdSigDmdEnv, hasDemandEnvSig,
- nopSig, botSig, isNopSig, isDeadEndSig, isDeadEndAppSig, trimBoxityDmdSig,
+ nopSig, botSig, isNopSig, isBottomingSig, isDeadEndSig, isDeadEndAppSig,
+ trimBoxityDmdSig, transferArgBoxityDmdSig,
-- ** Handling arity adjustments
prependArgsDmdSig, etaConvertDmdSig,
@@ -2147,6 +2148,13 @@ isNopSig (DmdSig ty) = isNopDmdType ty
isDeadEndSig :: DmdSig -> Bool
isDeadEndSig (DmdSig (DmdType _ _ res)) = isDeadEndDiv res
+-- | True if the signature diverges or throws an imprecise exception in a saturated call.
+-- NB: In constrast to 'isDeadEndSig' this returns False for 'exnDiv'.
+-- See Note [Dead ends]
+-- and Note [Precise vs imprecise exceptions].
+isBottomingSig :: DmdSig -> Bool
+isBottomingSig (DmdSig (DmdType _ _ res)) = res == botDiv
+
-- | True when the signature indicates all arguments are boxed
onlyBoxedArguments :: DmdSig -> Bool
onlyBoxedArguments (DmdSig (DmdType _ dmds _)) = all demandIsBoxed dmds
@@ -2179,6 +2187,38 @@ trimBoxityDmdType (DmdType fvs ds res) =
trimBoxityDmdSig :: DmdSig -> DmdSig
trimBoxityDmdSig = coerce trimBoxityDmdType
+-- | Transfers the boxity of the left arg to the demand structure of the right
+-- arg. This only makes sense if applied to new and old demands of the same
+-- value.
+transferBoxity :: Demand -> Demand -> Demand
+transferBoxity from to = go_dmd from to
+ where
+ go_dmd (from_n :* from_sd) to_dmd@(to_n :* to_sd)
+ | isAbs from_n || isAbs to_n = to_dmd
+ | otherwise = case (from_sd, to_sd) of
+ (Poly from_b _, Poly _ to_c) ->
+ to_n :* Poly from_b to_c
+ (_, Prod _ to_ds)
+ | Just (from_b, from_ds) <- viewProd (length to_ds) from_sd
+ -> to_n :* mkProd from_b (strictZipWith go_dmd from_ds to_ds)
+ (Prod from_b from_ds, _)
+ | Just (_, to_ds) <- viewProd (length from_ds) to_sd
+ -> to_n :* mkProd from_b (strictZipWith go_dmd from_ds to_ds)
+ _ -> trimBoxity to_dmd
+
+transferArgBoxityDmdType :: DmdType -> DmdType -> DmdType
+transferArgBoxityDmdType _from@(DmdType _ from_ds _) to@(DmdType to_fvs to_ds to_res)
+ | equalLength from_ds to_ds
+ = -- pprTraceWith "transfer" (\r -> ppr _from $$ ppr to $$ ppr r) $
+ DmdType to_fvs -- Only arg boxity! See Note [Don't change boxity without worker/wrapper]
+ (zipWith transferBoxity from_ds to_ds)
+ to_res
+ | otherwise
+ = trimBoxityDmdType to
+
+transferArgBoxityDmdSig :: DmdSig -> DmdSig -> DmdSig
+transferArgBoxityDmdSig = coerce transferArgBoxityDmdType
+
prependArgsDmdSig :: Int -> DmdSig -> DmdSig
-- ^ Add extra ('topDmd') arguments to a strictness signature.
-- In contrast to 'etaConvertDmdSig', this /prepends/ additional argument
=====================================
testsuite/tests/simplCore/should_compile/spec-inline.stderr
=====================================
@@ -143,7 +143,7 @@ Roman.foo1 = GHC.Maybe.Just @Int Roman.foo2
foo :: Int -> Int
[GblId,
Arity=1,
- Str=<1!P(L)>,
+ Str=<1L>,
Cpr=1,
Unf=Unf{Src=StableSystem, TopLvl=True, Value=True, ConLike=True,
WorkFree=True, Expandable=True,
=====================================
testsuite/tests/stranal/sigs/T21754.hs
=====================================
@@ -0,0 +1,7 @@
+{-# OPTIONS_GHC -fno-worker-wrapper #-}
+
+module Test where
+
+f :: Int -> Int
+f n = n+1
+{-# NOINLINE f #-}
=====================================
testsuite/tests/stranal/sigs/T21754.stderr
=====================================
@@ -0,0 +1,10 @@
+
+==================== Strictness signatures ====================
+Test.f: <1L>
+
+
+
+==================== Strictness signatures ====================
+Test.f: <1L>
+
+
=====================================
testsuite/tests/stranal/sigs/all.T
=====================================
@@ -34,5 +34,6 @@ test('T20746b', normal, compile, [''])
test('T21081', normal, compile, [''])
test('T21119', normal, compile, [''])
test('T21717', normal, compile, [''])
+test('T21754', normal, compile, [''])
test('T21888', normal, compile, [''])
test('T21888a', normal, compile, [''])
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/f5e8f493b015df859833beac5a8e64a0f9b9d4f4
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/f5e8f493b015df859833beac5a8e64a0f9b9d4f4
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/20220930/6787521c/attachment-0001.html>
More information about the ghc-commits
mailing list