| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Triggered by #12150, and the knock-on effects of join points, I did a
major refactoring of the Simplifier. This is a big patch that change
a lot of Simplify.hs: I did a lot of other re-organisation.
The main event
~~~~~~~~~~~~~~
Since the dawn of time we have had
simplExpr :: SimplEnv -> InExpr -> SimplCont
-> SimplM (SimplEnv, OutExpr)
What's that SimplEnv in the result? When simplifying an expression the
simplifier add floated let-bindings to the SimplEnv, extending the
in-scope set appropriately, and hence needs to resturn the SimplEnv at
the end. The mode, flags, substitution in the returned SimplEnv were
all irrelevant: it was just the floating bindings.
It's strange to accumulate part of the /result/ in the /environment/
argument! And indeed its leads to all manner of mysterious calls to
zapFloats and transferring of floats from one SimplEnv to another.
It got worse with join points, so I finally bit the bullet and refactored.
Now we have
simplExpr :: SimplEnv -> InExpr -> SimplCont
-> SimplM (SimplFloats, OutExpr)
-- See Note [The big picture]
and the SimplEnv no longer has floats in it. The code is no shorter,
but it /is/ easier to understand.
Main changes
* Remove seLetFloats field from SimplEnv
* Define new data type SimplFloats, and functions over it
* Change the types of simplExpr, simplBind, and their many variants,
to follow the above plan
Bottoming bindings
~~~~~~~~~~~~~~~~~~
I made one other significant change in SimplUtils (not just refactoring),
related to Trac #12150 comment:16. Given
x = <rhs>
where <rhs> turns out to be a bottoming expression, propagate that
information to x's IdInfo immediately. That's always good, because
it makes x be inlined less (we don't inline bottoming things), and
it allows (case x of ...) to drop the dead alterantives immediately.
Moreover, we are doing the analysis anyway, in tryEtaExpandRhs, which
calls CoreArity.findRhsArity, which already does simple bottom analysis.
So we are generating the information; all we need do is to atach the
bottoming info to the IdInfo.
See Note [Bottoming bindings]
Smaller refactoring
~~~~~~~~~~~~~~~~~~~
* Rename SimplifierMode to SimplMode
* Put DynFlags as a new field in SimplMode, to make fewer
monadic calls to getDynFlags.
* Move the code in addPolyBind into abstractFloats
* Move the "don't eta-expand join points" into tryEtaExpandRhs
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This very small patch switches on sm_inline even in the InitialPhase
(aka "gentle" phase). There is no reason not to... and the results
are astonishing.
I think the peformance of GHC itself improves by about 5%; and some
programs get much smaller, quicker. Result: across the board
irmprovements in
compile time performance. Here are the changes in perf/compiler;
the numbers are decreases in compiler bytes-allocated:
3% T5837
7% parsing001
9% T12234
35% T9020
9% T3064
13% T9961
20% T13056
5% T9872d
5% T9872c
5% T9872b
7% T9872a
5% T783
35% T12227
20% T1969
Plus in perf/should_run
5% lazy-bs-alloc
It wasn't as easy as it sounds: I did a raft of preparatory work in
earlier patches. But it's great!
Reviewers: austin, bgamari
Subscribers: thomie
Differential Revision: https://phabricator.haskell.org/D3203
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commits relaxes the invariants of the Core syntax so that a
top-level variable can be bound to a primitive string literal of type
Addr#.
This commit:
* Relaxes the invatiants of the Core, and allows top-level bindings whose
type is Addr# as long as their RHS is either a primitive string literal or
another variable.
* Allows the simplifier and the full-laziness transformer to float out
primitive string literals to the top leve.
* Introduces the new StgGenTopBinding type to accomodate top-level Addr#
bindings.
* Introduces a new type of labels in the object code, with the suffix "_bytes",
for exported top-level Addr# bindings.
* Makes some built-in rules more robust. This was necessary to keep them
functional after the above changes.
This is a continuation of D2554.
Rebasing notes:
This had two slightly suspicious performance regressions:
* T12425: bytes allocated regressed by roughly 5%
* T4029: bytes allocated regressed by a bit over 1%
* T13035: bytes allocated regressed by a bit over 5%
These deserve additional investigation.
Rebased by: bgamari.
Test Plan: ./validate --slow
Reviewers: goldfire, trofi, simonmar, simonpj, austin, hvr, bgamari
Reviewed By: trofi, simonpj, bgamari
Subscribers: trofi, simonpj, gridaphobe, thomie
Differential Revision: https://phabricator.haskell.org/D2605
GHC Trac Issues: #8472
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This implements the ideas originally put forward in
"System FC with Explicit Kind Equality" (ICFP'13).
There are several noteworthy changes with this patch:
* We now have casts in types. These change the kind
of a type. See new constructor `CastTy`.
* All types and all constructors can be promoted.
This includes GADT constructors. GADT pattern matches
take place in type family equations. In Core,
types can now be applied to coercions via the
`CoercionTy` constructor.
* Coercions can now be heterogeneous, relating types
of different kinds. A coercion proving `t1 :: k1 ~ t2 :: k2`
proves both that `t1` and `t2` are the same and also that
`k1` and `k2` are the same.
* The `Coercion` type has been significantly enhanced.
The documentation in `docs/core-spec/core-spec.pdf` reflects
the new reality.
* The type of `*` is now `*`. No more `BOX`.
* Users can write explicit kind variables in their code,
anywhere they can write type variables. For backward compatibility,
automatic inference of kind-variable binding is still permitted.
* The new extension `TypeInType` turns on the new user-facing
features.
* Type families and synonyms are now promoted to kinds. This causes
trouble with parsing `*`, leading to the somewhat awkward new
`HsAppsTy` constructor for `HsType`. This is dispatched with in
the renamer, where the kind `*` can be told apart from a
type-level multiplication operator. Without `-XTypeInType` the
old behavior persists. With `-XTypeInType`, you need to import
`Data.Kind` to get `*`, also known as `Type`.
* The kind-checking algorithms in TcHsType have been significantly
rewritten to allow for enhanced kinds.
* The new features are still quite experimental and may be in flux.
* TODO: Several open tickets: #11195, #11196, #11197, #11198, #11203.
* TODO: Update user manual.
Tickets addressed: #9017, #9173, #7961, #10524, #8566, #11142.
Updates Haddock submodule.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch began as a modest refactoring of HsType and friends, to
clarify and tidy up exactly where quantification takes place in types.
Although initially driven by making the implementation of wildcards more
tidy (and fixing a number of bugs), I gradually got drawn into a pretty
big process, which I've been doing on and off for quite a long time.
There is one compiler performance regression as a result of all
this, in perf/compiler/T3064. I still need to look into that.
* The principal driving change is described in Note [HsType binders]
in HsType. Well worth reading!
* Those data type changes drive almost everything else. In particular
we now statically know where
(a) implicit quantification only (LHsSigType),
e.g. in instance declaratios and SPECIALISE signatures
(b) implicit quantification and wildcards (LHsSigWcType)
can appear, e.g. in function type signatures
* As part of this change, HsForAllTy is (a) simplified (no wildcards)
and (b) split into HsForAllTy and HsQualTy. The two contructors
appear when and only when the correponding user-level construct
appears. Again see Note [HsType binders].
HsExplicitFlag disappears altogether.
* Other simplifications
- ExprWithTySig no longer needs an ExprWithTySigOut variant
- TypeSig no longer needs a PostRn name [name] field
for wildcards
- PatSynSig records a LHsSigType rather than the decomposed
pieces
- The mysterious 'GenericSig' is now 'ClassOpSig'
* Renamed LHsTyVarBndrs to LHsQTyVars
* There are some uninteresting knock-on changes in Haddock,
because of the HsSyn changes
I also did a bunch of loosely-related changes:
* We already had type synonyms CoercionN/CoercionR for nominal and
representational coercions. I've added similar treatment for
TcCoercionN/TcCoercionR
mkWpCastN/mkWpCastN
All just type synonyms but jolly useful.
* I record-ised ForeignImport and ForeignExport
* I improved the (poor) fix to Trac #10896, by making
TcTyClsDecls.checkValidTyCl recover from errors, but adding a
harmless, abstract TyCon to the envt if so.
* I did some significant refactoring in RnEnv.lookupSubBndrOcc,
for reasons that I have (embarrassingly) now totally forgotten.
It had to do with something to do with import and export
Updates haddock submodule.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is the second attempt at merging D757.
This patch implements the idea floated in Trac #9858, namely that we
should generate type-representation information at the data type
declaration site, rather than when solving a Typeable constraint.
However, this turned out quite a bit harder than I expected. I still
think it's the right thing to do, and it's done now, but it was quite
a struggle.
See particularly
* Note [Grand plan for Typeable] in TcTypeable (which is a new module)
* Note [The overall promotion story] in DataCon (clarifies existing
stuff)
The most painful bit was that to generate Typeable instances (ie
TyConRepName bindings) for every TyCon is tricky for types in ghc-prim
etc:
* We need to have enough data types around to *define* a TyCon
* Many of these types are wired-in
Also, to minimise the code generated for each data type, I wanted to
generate pure data, not CAFs with unpackCString# stuff floating about.
Performance
~~~~~~~~~~~
Three perf/compiler tests start to allocate quite a bit more. This isn't
surprising, because they all allocate zillions of data types, with
practically no other code, esp. T1969
* T1969: GHC allocates 19% more
* T4801: GHC allocates 13% more
* T5321FD: GHC allocates 13% more
* T9675: GHC allocates 11% more
* T783: GHC allocates 11% more
* T5642: GHC allocates 10% more
I'm treating this as acceptable. The payoff comes in Typeable-heavy
code.
Remaining to do
~~~~~~~~~~~~~~~
* I think that "TyCon" and "Module" are over-generic names to use for
the runtime type representations used in GHC.Typeable. Better might
be
"TrTyCon" and "TrModule". But I have not yet done this
* Add more info the the "TyCon" e.g. source location where it was
defined
* Use the new "Module" type to help with Trac Trac #10068
* It would be possible to generate TyConRepName (ie Typeable
instances) selectively rather than all the time. We'd need to persist
the information in interface files. Lacking a motivating reason I
have
not done this, but it would not be difficult.
Refactoring
~~~~~~~~~~~
As is so often the case, I ended up refactoring more than I intended.
In particular
* In TyCon, a type *family* (whether type or data) is repesented by a
FamilyTyCon
* a algebraic data type (including data/newtype instances) is
represented by AlgTyCon This wasn't true before; a data family
was represented as an AlgTyCon. There are some corresponding
changes in IfaceSyn.
* Also get rid of the (unhelpfully named) tyConParent.
* In TyCon define 'Promoted', isomorphic to Maybe, used when things are
optionally promoted; and use it elsewhere in GHC.
* Cleanup handling of knownKeyNames
* Each TyCon, including promoted TyCons, contains its TyConRepName, if
it has one. This is, in effect, the name of its Typeable instance.
Updates haddock submodule
Test Plan: Let Harbormaster validate
Reviewers: austin, hvr, goldfire
Subscribers: goldfire, thomie
Differential Revision: https://phabricator.haskell.org/D1404
GHC Trac Issues: #9858
|
| |
|
|
|
|
|
|
| |
This reverts commit bef2f03e4d56d88a7e9752a7afd6a0a35616da6c.
This merge was botched
Also reverts haddock submodule.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch implements the idea floated in Trac #9858, namely that we
should generate type-representation information at the data type
declaration site, rather than when solving a Typeable constraint.
However, this turned out quite a bit harder than I expected. I still
think it's the right thing to do, and it's done now, but it was quite
a struggle.
See particularly
* Note [Grand plan for Typeable] in TcTypeable (which is a new module)
* Note [The overall promotion story] in DataCon (clarifies existing stuff)
The most painful bit was that to generate Typeable instances (ie
TyConRepName bindings) for every TyCon is tricky for types in ghc-prim
etc:
* We need to have enough data types around to *define* a TyCon
* Many of these types are wired-in
Also, to minimise the code generated for each data type, I wanted to
generate pure data, not CAFs with unpackCString# stuff floating about.
Performance
~~~~~~~~~~~
Three perf/compiler tests start to allocate quite a bit more. This isn't
surprising, because they all allocate zillions of data types, with
practically no other code, esp. T1969
* T3294: GHC allocates 110% more (filed #11030 to track this)
* T1969: GHC allocates 30% more
* T4801: GHC allocates 14% more
* T5321FD: GHC allocates 13% more
* T783: GHC allocates 12% more
* T9675: GHC allocates 12% more
* T5642: GHC allocates 10% more
* T9961: GHC allocates 6% more
* T9203: Program allocates 54% less
I'm treating this as acceptable. The payoff comes in Typeable-heavy
code.
Remaining to do
~~~~~~~~~~~~~~~
* I think that "TyCon" and "Module" are over-generic names to use for
the runtime type representations used in GHC.Typeable. Better might be
"TrTyCon" and "TrModule". But I have not yet done this
* Add more info the the "TyCon" e.g. source location where it was
defined
* Use the new "Module" type to help with Trac Trac #10068
* It would be possible to generate TyConRepName (ie Typeable
instances) selectively rather than all the time. We'd need to persist
the information in interface files. Lacking a motivating reason I have
not done this, but it would not be difficult.
Refactoring
~~~~~~~~~~~
As is so often the case, I ended up refactoring more than I intended.
In particular
* In TyCon, a type *family* (whether type or data) is repesented by a
FamilyTyCon
* a algebraic data type (including data/newtype instances) is
represented by AlgTyCon This wasn't true before; a data family
was represented as an AlgTyCon. There are some corresponding
changes in IfaceSyn.
* Also get rid of the (unhelpfully named) tyConParent.
* In TyCon define 'Promoted', isomorphic to Maybe, used when things are
optionally promoted; and use it elsewhere in GHC.
* Cleanup handling of knownKeyNames
* Each TyCon, including promoted TyCons, contains its TyConRepName, if
it has one. This is, in effect, the name of its Typeable instance.
Requires update of the haddock submodule.
Differential Revision: https://phabricator.haskell.org/D757
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This long-standing, terrible, adn somewhat subtle bug was exposed
by Trac #10370, thanks to Reid Barton's brilliant test case (comment:3).
The effect is large on the Trac #10370 test.
Here is what the profile report says:
Before:
total time = 24.35 secs (24353 ticks @ 1000 us, 1 processor)
total alloc = 11,864,360,816 bytes (excludes profiling overheads)
After:
total time = 21.16 secs (21160 ticks @ 1000 us, 1 processor)
total alloc = 7,947,141,136 bytes (excludes profiling overheads)
The /combined/ effect of the tidyOccName fix, plus this one, is dramtic
for Trac #10370. Here is what +RTS -s says:
Before:
15,490,210,952 bytes allocated in the heap
1,783,919,456 bytes maximum residency (20 sample(s))
MUT time 30.117s ( 31.383s elapsed)
GC time 90.103s ( 90.107s elapsed)
Total time 120.843s (122.065s elapsed)
After:
7,928,671,936 bytes allocated in the heap
52,914,832 bytes maximum residency (25 sample(s))
MUT time 13.912s ( 15.110s elapsed)
GC time 6.809s ( 6.808s elapsed)
Total time 20.789s ( 21.954s elapsed)
- Heap allocation halved
- Residency cut by a factor of more than 30.
- ELapsed time cut by a factor of 6
Not bad!
The details
~~~~~~~~~~~
The culprit was SimplEnv.mkCoreSubst, which used mapVarEnv to do some
impedence-matching from the substitituion used by the simplifier to
the one used by CoreSubst. But the impedence-mactching was recursive!
mk_subst tv_env cv_env id_env
= CoreSubst.mkSubst in_scope tv_env cv_env (mapVarEnv fiddle id_env)
fiddle (DoneEx e) = e
fiddle (DoneId v) = Var v
fiddle (ContEx tv cv id e) = CoreSubst.substExpr (mk_subst tv cv id) e
Inside fiddle, in the ContEx case, we may do another whole level of
fiddle. And so on. Moreover, UniqFM (which is built on Data.IntMap) is
strict, so the fiddling is done eagerly. I didn't wok through all the
details but the result is a gargatuan blow-up of entirely unnecessary work.
Laziness would make this go away, I think, but I don't want to mess
with IntMap. And in any case, the impedence matching is a royal pain.
In the end I simply ceased trying to use CoreSubst.substExpr in the
simplifier, and instead just use simplExpr. That does mean bit of
duplication; e.g. new code for simplRules. But it's not a big deal
and it's far more direct and easy to reason about.
A bit of knock-on refactoring:
* Data type ArgSummary moves to CoreUnfold.
* interestingArg moves from CoreUnfold to SimplUtils, and gets a
SimplEnv argument which can be used when we encounter a variable.
* simplLamBndrs, addBndrRules move from SimplEnv to Simplify
(because they now calls simplUnfolding, simplRules resp)
* SimplUtils.substExpr, substUnfolding, mkCoreSubst die completely
* In Simplify some several functions that were previously pure
substitution-based functions are now monadic:
- addBndrRules, simplRule
- addCoerce, add_coerce in simplCast
* In case 2c of Simplify.rebuildCase, there was a pretty disgusting
expression-substitution taking place for 'rhs'; and we really don't
want to make that monadic becuase 'rhs' can be big.
Solution: reduce the arity of the rules for seq.
See Note [User-defined RULES for seq] in MkId.
|
| | |
|
| |
|
|
|
|
|
|
| |
Some nice improvements on already succeeding test cases (#876, #7954
and #4267)
Test #149 needed a little change, lest call arity causes a allocation
change that we do not want to test here.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
| |
This big patch implements a kind-polymorphic core for GHC. The current
implementation focuses on making sure that all kind-monomorphic programs still
work in the new core; it is not yet guaranteed that kind-polymorphic programs
(using the new -XPolyKinds flag) will work.
For more information, see http://haskell.org/haskellwiki/GHC/Kinds
|
| |
|
