| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Thanks to ghc-bignum, the compiler can be simplified:
* Types and constructors of Integer and Natural can be wired-in. It
means that we don't have to query them from interfaces. It also means
that numeric literals don't have to carry their type with them.
* The same code is used whatever ghc-bignum backend is enabled. In
particular, conversion of bignum literals into final Core expressions
is now much more straightforward. Bignum closure inspection too.
* GHC itself doesn't depend on any integer-* package anymore
* The `integerLibrary` setting is gone.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This implements several general performance improvements to GHC,
to offset the effect of the linear types change.
General optimisations:
- Add a `coreFullView` function which iterates `coreView` on the
head. This avoids making function recursive solely because the
iterate `coreView` themselves. As a consequence, this functions can
be inlined, and trigger case-of-known constructor (_e.g._
`kindRep_maybe`, `isLiftedRuntimeRep`, `isMultiplicityTy`,
`getTyVar_maybe`, `splitAppTy_maybe`, `splitFunType_maybe`,
`tyConAppTyCon_maybe`). The common pattern about all these functions
is that they are almost always used as views, and immediately
consumed by a case expression. This commit also mark them asx `INLINE`.
- In `subst_ty` add a special case for nullary `TyConApp`, which avoid
allocations altogether.
- Use `mkTyConApp` in `subst_ty` for the general `TyConApp`. This
required quite a bit of module shuffling.
case. `myTyConApp` enforces crucial sharing, which was lost during
substitution. See also !2952 .
- Make `subst_ty` stricter.
- In `eqType` (specifically, in `nonDetCmpType`), add a special case,
tested first, for the very common case of nullary `TyConApp`.
`nonDetCmpType` has been made `INLINE` otherwise it is actually a
regression. This is similar to the optimisations in !2952.
Linear-type specific optimisations:
- Use `tyConAppTyCon_maybe` instead of the more complex `eqType` in
the definition of the pattern synonyms `One` and `Many`.
- Break the `hs-boot` cycles between `Multiplicity.hs` and `Type.hs`:
`Multiplicity` now import `Type` normally, rather than from the
`hs-boot`. This way `tyConAppTyCon_maybe` can inline properly in the
`One` and `Many` pattern synonyms.
- Make `updateIdTypeAndMult` strict in its type and multiplicity
- The `scaleIdBy` gets a specialised definition rather than being an
alias to `scaleVarBy`
- `splitFunTy_maybe` is given the type `Type -> Maybe (Mult, Type,
Type)` instead of `Type -> Maybe (Scaled Type, Type)`
- Remove the `MultMul` pattern synonym in favour of a view `isMultMul`
because pattern synonyms appear not to inline well.
- in `eqType`, in a `FunTy`, compare multiplicities last: they are
almost always both `Many`, so it helps failing faster.
- Cache `manyDataConTy` in `mkTyConApp`, to make sure that all the
instances of `TyConApp ManyDataConTy []` are physically the same.
This commit has been authored by
* Richard Eisenberg
* Krzysztof Gogolewski
* Arnaud Spiwack
Metric Decrease:
haddock.base
T12227
T12545
T12990
T1969
T3064
T5030
T9872b
Metric Increase:
haddock.base
haddock.Cabal
haddock.compiler
T12150
T12234
T12425
T12707
T13035
T13056
T15164
T16190
T18304
T1969
T3064
T3294
T5631
T5642
T5837
T6048
T9020
T9233
T9675
T9872a
T9961
WWRec
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is the first step towards implementation of the linear types proposal
(https://github.com/ghc-proposals/ghc-proposals/pull/111).
It features
* A language extension -XLinearTypes
* Syntax for linear functions in the surface language
* Linearity checking in Core Lint, enabled with -dlinear-core-lint
* Core-to-core passes are mostly compatible with linearity
* Fields in a data type can be linear or unrestricted; linear fields
have multiplicity-polymorphic constructors.
If -XLinearTypes is disabled, the GADT syntax defaults to linear fields
The following items are not yet supported:
* a # m -> b syntax (only prefix FUN is supported for now)
* Full multiplicity inference (multiplicities are really only checked)
* Decent linearity error messages
* Linear let, where, and case expressions in the surface language
(each of these currently introduce the unrestricted variant)
* Multiplicity-parametric fields
* Syntax for annotating lambda-bound or let-bound with a multiplicity
* Syntax for non-linear/multiple-field-multiplicity records
* Linear projections for records with a single linear field
* Linear pattern synonyms
* Multiplicity coercions (test LinearPolyType)
A high-level description can be found at
https://ghc.haskell.org/trac/ghc/wiki/LinearTypes/Implementation
Following the link above you will find a description of the changes made to Core.
This commit has been authored by
* Richard Eisenberg
* Krzysztof Gogolewski
* Matthew Pickering
* Arnaud Spiwack
With contributions from:
* Mark Barbone
* Alexander Vershilov
Updates haddock submodule.
|
| |
|
|
|
|
|
| |
It avoids using DynFlags in the Outputable instance of Clabel to check
assertions at pretty-printing time.
|
|
|
|
|
| |
This bit of documentation got outdated after commit
1fcede43d2b30f33b7505e25eb6b1f321be0407f
|
| |
|
|
|
|
|
|
|
|
|
|
| |
The Haskell calling convention requires integer parameters smaller
than wordsize to be promoted to wordsize (where the upper bits are
don't care). To access such small integer parameter read a word from
the parameter array and then cast that word to the small integer
target type.
Fixes #15933
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously CoreToStg would unconditionally discard cases of the form:
case unsafeEqualityProof of wild { _ -> rhs }
and rather replace the whole thing with `rhs`. However, in some cases
(see #18227) the case binder is still live, resulting in unbound
occurrences in `rhs`. Fix this by only discarding the case if the case
binder is dead.
Fixes #18227.
|
|
|
|
|
|
|
| |
The initial version was rewritten by Tamar Christina.
It was rewritten in large parts by Andreas Klebinger.
Co-authored-by: Andreas Klebinger <klebinger.andreas@gmx.at>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, `HsForAllTy` permits the combination of `ForallVis` and
`Inferred`, but you can't actually typecheck code that uses it
(e.g., `forall {a} ->`). This patch refactors `HsForAllTy` to use a
new `HsForAllTelescope` data type that makes a type-level distinction
between visible and invisible `forall`s such that visible `forall`s
do not track `Specificity`. That part of the patch is actually quite
small; the rest is simply changing consumers of `HsType` to
accommodate this new type.
Fixes #18235. Bumps the `haddock` submodule.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Ticket #18304 showed that we need to be very careful
when exploring the demand (esp usage demand) on recursive
product types.
This patch solves the problem by trimming the demand on such types --
in effect, a form of "widening".
See the Note [Trimming a demand to a type] in DmdAnal, which explains
how I did this by piggy-backing on an existing mechansim for trimming
demands becuase of GADTs. The significant payload of this patch is
very small indeed:
* Make GHC.Core.Opt.WorkWrap.Utils.typeShape use RecTcChecker to
avoid looking through recursive types.
But on the way
* I found that ae_rec_tc was entirely inoperative and did nothing.
So I removed it altogether from DmdAnal.
* I moved some code around in DmdAnal and Demand.
(There are no actual changes in dmdFix.)
* I changed the API of DmsAnal.dmdAnalRhsLetDown to return
a StrictSig rather than a decorated Id
* I removed the dead function peelTsFuns from Demand
Performance effects:
Nofib: 0.0% changes. Not surprising, because they don't
use recursive products
Perf tests
T12227:
1% increase in compiler allocation, becuase $cto gets w/w'd.
It did not w/w before because it takes a deeply nested
argument, so the worker gets too many args, so we abandon w/w
altogether (see GHC.Core.Opt.WorkWrap.Utils.isWorkerSmallEnough)
With this patch we trim the demands. That is not strictly
necessary (since these Generic type constructors are like
tuples -- they can't cause a loop) but the net result is that
we now w/w $cto which is fine.
UniqLoop:
16% decrease in /runtime/ allocation. The UniqSupply is a
recursive product, so currently we abandon all strictness on
'churn'. With this patch 'churn' gets useful strictness, and
we w/w it. Hooray
Metric Decrease:
UniqLoop
Metric Increase:
T12227
|
| |
|
| |
|
|
|
|
| |
Preload units can be retrieved in UnitState when needed (i.e. in GHCi)
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
Avoid directly querying flags from DynFlags to build the UnitState.
Instead go via UnitConfig so that we could reuse this to make another
UnitState for plugins.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
* Remove preload parameter (unused)
* Don't explicitly return preloaded units: redundant because already
returned as "preloadUnits" field of UnitState
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
| |
* rename PackageState into UnitState
* rename findWiredInPackages into findWiredInUnits
* rename lookupModuleInAll[Packages,Units]
* etc.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The terminology changed over time and now package databases contain
"units" (there can be several units compiled from a single Cabal
package: one per-component, one for each option set, one per
instantiation, etc.). We should try to be consistent internally and use
"units": that's what this renaming does. Maybe one day we'll fix the UI
too (e.g. replace -package-id with -unit-id, we already have
-this-unit-id and ghc-pkg has -unit-id...) but it's not done in this
patch.
* rename getPkgFrameworkOpts into getUnitFrameworkOpts
* rename UnitInfoMap into ClosureUnitInfoMap
* rename InstalledPackageIndex into UnitInfoMap
* rename UnusablePackages into UnusableUnits
* rename PackagePrecedenceIndex into UnitPrecedenceMap
* rename PackageDatabase into UnitDatabase
* rename pkgDatabase into unitDatabases
* rename pkgState into unitState
* rename initPackages into initUnits
* rename renamePackage into renameUnitInfo
* rename UnusablePackageReason into UnusableUnitReason
* rename getPackage* into getUnit*
* etc.
|
|
|
|
| |
There is no Map involved
|
| |
|
| |
|
|
|
|
| |
* Remove WiredInUnitId and WiredUnitId type aliases
|
| |
|
|
|
|
|
|
|
|
| |
* use UnitId instead of String to identify wired-in units
* use UnitId instead of Unit in the backend (Unit are only use by
Backpack to produce type-checked interfaces, not real code)
* rename lookup functions for consistency
* documentation
|
|
|
|
|
|
|
|
|
|
| |
Instead of always querying DynFlags to know whether we are allowed to
use virtual units (i.e. instantiated on-the-fly, cf Note [About units]
in GHC.Unit), we store it once for all in
`PackageState.allowVirtualUnits`.
This avoids using DynFlags too much (cf #17957) and is preliminary work
for #14335.
|
| |
|
|
|
|
|
| |
* rename thisPackage into homeUnit
* document and refactor several Backpack things
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
* Remove several uses of `sdocWithDynFlags`, especially in GHC.Llvm.*
* Add LlvmOpts datatype to store Llvm backend options
* Remove Outputable instances (for LlvmVar, LlvmLit, LlvmStatic and
Llvm.MetaExpr) which require LlvmOpts.
* Rename ppMetaExpr into ppMetaAnnotExpr (pprMetaExpr is now used in place of `ppr :: MetaExpr -> SDoc`)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In some cases it was possible for lookupGlobalOccRn_maybe to return an
error, when it should be returning a Nothing. If it called
lookupExactOcc_either when there were no matching GlobalRdrElts in the
otherwise case, it would return an error message. This could be caused
when lookupThName_maybe in Template Haskell was looking in different
namespaces (thRdrNameGuesses), guessing different namespaces that the
name wasn't guaranteed to be found in.
However, by addressing this some more accurate errors were being lost in
the conversion to Maybes. So some of the lookup* functions have been
shuffled about so that errors should always be ignored in
lookup*_maybes, and propagated otherwise.
This fixes #18263
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The cast worker/wrapper transformation transforms
x = e |> co
into
y = e
x = y |> co
This is done by the simplifier, but we were being
careless about transferring IdInfo from x to y,
and about what to do if x is a NOINLNE function.
This resulted in a series of bugs:
#17673, #18093, #18078.
This patch fixes all that:
* Main change is in GHC.Core.Opt.Simplify, and
the new prepareBinding function, which does this
cast worker/wrapper transform.
See Note [Cast worker/wrappers].
* There is quite a bit of refactoring around
prepareRhs, makeTrivial etc. It's nicer now.
* Some wrappers from strictness and cast w/w, notably those for
a function with a NOINLINE, should inline very late. There
wasn't really a mechanism for that, which was an existing bug
really; so I invented a new finalPhase = Phase (-1). It's used
for all simplifier runs after the user-visible phase 2,1,0 have
run. (No new runs of the simplifier are introduced thereby.)
See new Note [Compiler phases] in GHC.Types.Basic;
the main changes are in GHC.Core.Opt.Driver
* Doing this made me trip over two places where the AnonArgFlag on a
FunTy was being lost so we could end up with (Num a -> ty)
rather than (Num a => ty)
- In coercionLKind/coercionRKind
- In contHoleType in the Simplifier
I fixed the former by defining mkFunctionType and using it in
coercionLKind/RKind.
I could have done the same for the latter, but the information
is almost to hand. So I fixed the latter by
- adding sc_hole_ty to ApplyToVal (like ApplyToTy),
- adding as_hole_ty to ValArg (like TyArg)
- adding sc_fun_ty to StrictArg
Turned out I could then remove ai_type from ArgInfo. This is
just moving the deck chairs around, but it worked out nicely.
See the new Note [AnonArgFlag] in GHC.Types.Var
* When looking at the 'arity decrease' thing (#18093) I discovered
that stable unfoldings had a much lower arity than the actual
optimised function. That's what led to the arity-decrease
message. Simple solution: eta-expand.
It's described in Note [Eta-expand stable unfoldings]
in GHC.Core.Opt.Simplify
* I also discovered that unsafeCoerce wasn't being inlined if
the context was boring. So (\x. f (unsafeCoerce x)) would
create a thunk -- yikes! I fixed that by making inlineBoringOK
a bit cleverer: see Note [Inline unsafeCoerce] in GHC.Core.Unfold.
I also found that unsafeCoerceName was unused, so I removed it.
I made a test case for #18078, and a very similar one for #17673.
The net effect of all this on nofib is very modest, but positive:
--------------------------------------------------------------------------------
Program Size Allocs Runtime Elapsed TotalMem
--------------------------------------------------------------------------------
anna -0.4% -0.1% -3.1% -3.1% 0.0%
fannkuch-redux -0.4% -0.3% -0.1% -0.1% 0.0%
maillist -0.4% -0.1% -7.8% -1.0% -14.3%
primetest -0.4% -15.6% -7.1% -6.6% 0.0%
--------------------------------------------------------------------------------
Min -0.9% -15.6% -13.3% -14.2% -14.3%
Max -0.3% 0.0% +12.1% +12.4% 0.0%
Geometric Mean -0.4% -0.2% -2.3% -2.2% -0.1%
All following metric decreases are compile-time allocation decreases
between -1% and -3%:
Metric Decrease:
T5631
T13701
T14697
T15164
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch switches on -fno-state-hack in GHC.Types.Unique.Supply.
It turned out that my fixes for #18078 (coercion floating) changed the
optimisation pathway for mkSplitUniqSupply in such a way that we had
an extra allocation inside the inner loop. Adding -fno-state-hack
fixed that -- and indeed the loop in mkSplitUniqSupply is a classic
example of the way in which -fno-state-hack can be bad; see #18238.
Moreover, the new code is better than the old. They allocate
the same, but the old code ends up with a partial application.
The net effect is that the test
perf/should_run/UniqLoop
runs 20% faster! From 2.5s down to 2.0s. The allocation numbers
are the same -- but elapsed time falls. Good!
The bad thing about this is that it's terribly delicate. But
at least it's a good example of such delicacy in action.
There is a long Note [Optimising the unique supply] which now
explains all this.
|