| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The naive way of putting out n characters of indent would be something
like `hPutStr hdl (replicate n ' ')`. However this is quite inefficient
as we allocate an absurd number of strings consisting of simply spaces
as we don't cache them.
To improve on this we now track if we can simply write ascii spaces via
hPutBuf instead. This is the case when running with -ddump-to-file where
we force the encoding to be UTF8.
This avoids both the cost of going through encoding as well as avoiding
allocation churn from all the white space. Instead we simply use hPutBuf
on a preallocated unlifted string.
When dumping stg like this:
> nofib/spectral/simple/Main.hs -fforce-recomp -ddump-stg-final -ddump-to-file -c +RTS -s
Allocations went from 1,778 MB to 1,702MB. About a 4% reduction of
allocation! I did not measure the difference in runtime but expect it
to be similar.
Bumps the haddock submodule since the interface of GHC's Pretty
slightly changed.
-------------------------
Metric Decrease:
T12227
-------------------------
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Sinking requires us to track live local regs after each
cmm statement. We used to do this via "Set LocalReg".
However we can replace this with a solution based on IntSet
which is overall more efficient without losing much. The thing
we lose is width of the variables, which isn't used by the sinking
pass anyway.
I also reworked how we keep assignments to regs mentioned in
skipped assignments. I put the details into
Note [Keeping assignemnts mentioned in skipped RHSs].
The gist of it is instead of keeping track of it via the use count
which is a `IntMap Int` we now use the live regs set (IntSet) which
is quite a bit faster.
I think it also matches the semantics a lot better. The skipped
(not discarded) assignment does in fact keep the regs on it's rhs
alive so keeping track of this in the live set seems like the clearer
solution as well.
Improves allocations for T3294 by yet another 1%.
|
|
|
|
|
|
|
|
| |
Now that flattening doesn't produce flattening variables,
it's not really flattening anything: it's rewriting. This
change also means that the rewriter can no longer be confused
the core flattener (in GHC.Core.Unify), which is sometimes used
during type-checking.
|
|
|
|
|
|
|
| |
This will allow us to back out the allocations per compiler pass from
the eventlog. Note that we dump the allocation counter rather than the
difference since this will allow us to determine how much work is done
*between* `withTiming` blocks.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This replaces all Word<N> = W<N># Word# and Int<N> = I<N># Int# with
Word<N> = W<N># Word<N># and Int<N> = I<N># Int<N>#, thus providing us
with properly sized primitives in the codegenerator instead of pretending
they are all full machine words.
This came up when implementing darwinpcs for arm64. The darwinpcs reqires
us to pack function argugments in excess of registers on the stack. While
most procedure call standards (pcs) assume arguments are just passed in
8 byte slots; and thus the caller does not know the exact signature to make
the call, darwinpcs requires us to adhere to the prototype, and thus have
the correct sizes. If we specify CInt in the FFI call, it should correspond
to the C int, and not just be Word sized, when it's only half the size.
This does change the expected output of T16402 but the new result is no
less correct as it eliminates the narrowing (instead of the `and` as was
previously done).
Bumps the array, bytestring, text, and binary submodules.
Co-Authored-By: Ben Gamari <ben@well-typed.com>
Metric Increase:
T13701
T14697
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As outlined in #18903, interleaving usage and strictness demands not
only means a more compact demand representation, but also allows us to
express demands that we weren't easily able to express before.
Call demands are *relative* in the sense that a call demand `Cn(cd)`
on `g` says "`g` is called `n` times. *Whenever `g` is called*, the
result is used according to `cd`". Example from #18903:
```hs
h :: Int -> Int
h m =
let g :: Int -> (Int,Int)
g 1 = (m, 0)
g n = (2 * n, 2 `div` n)
{-# NOINLINE g #-}
in case m of
1 -> 0
2 -> snd (g m)
_ -> uncurry (+) (g m)
```
Without the interleaved representation, we would just get `L` for the
strictness demand on `g`. Now we are able to express that whenever
`g` is called, its second component is used strictly in denoting `g`
by `1C1(P(1P(U),SP(U)))`. This would allow Nested CPR to unbox the
division, for example.
Fixes #18903.
While fixing regressions, I also discovered and fixed #18957.
Metric Decrease:
T13253-spj
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This refactors the GHC AST to remove `HsImplicitBndrs` and replace it with
`HsOuterTyVarBndrs`, a type which records whether the outermost quantification
in a type is explicit (i.e., with an outermost, invisible `forall`) or
implicit. As a result of this refactoring, it is now evident in the AST where
the `forall`-or-nothing rule applies: it's all the places that use
`HsOuterTyVarBndrs`. See the revamped `Note [forall-or-nothing rule]` in
`GHC.Hs.Type` (previously in `GHC.Rename.HsType`).
Moreover, the places where `ScopedTypeVariables` brings lexically scoped type
variables into scope are a subset of the places that adhere to the
`forall`-or-nothing rule, so this also makes places that interact with
`ScopedTypeVariables` easier to find. See the revamped
`Note [Lexically scoped type variables]` in `GHC.Hs.Type` (previously in
`GHC.Tc.Gen.Sig`).
`HsOuterTyVarBndrs` are used in type signatures (see `HsOuterSigTyVarBndrs`)
and type family equations (see `HsOuterFamEqnTyVarBndrs`). The main difference
between the former and the latter is that the former cares about specificity
but the latter does not.
There are a number of knock-on consequences:
* There is now a dedicated `HsSigType` type, which is the combination of
`HsOuterSigTyVarBndrs` and `HsType`. `LHsSigType` is now an alias for an
`XRec` of `HsSigType`.
* Working out the details led us to a substantial refactoring of
the handling of explicit (user-written) and implicit type-variable
bindings in `GHC.Tc.Gen.HsType`.
Instead of a confusing family of higher order functions, we now
have a local data type, `SkolemInfo`, that controls how these
binders are kind-checked.
It remains very fiddly, not fully satisfying. But it's better
than it was.
Fixes #16762. Bumps the Haddock submodule.
Co-authored-by: Simon Peyton Jones <simonpj@microsoft.com>
Co-authored-by: Richard Eisenberg <rae@richarde.dev>
Co-authored-by: Zubin Duggal <zubin@cmi.ac.in>
|
| |
|
|
|
|
|
|
|
| |
While, say, alternating "he" and "she" in sequential writing
may be nicer than always using "they", reading code/documentation
is almost never sequential. If this small change makes individuals
feel more welcome in GHC's codebase, that's a good thing.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
I was working on making DynFlags stateless (#17957), especially by
storing loaded plugins into HscEnv instead of DynFlags. It turned out to
be complicated because HscEnv is in GHC.Driver.Types but LoadedPlugin
isn't: it is in GHC.Driver.Plugins which depends on GHC.Driver.Types. I
didn't feel like introducing yet another hs-boot file to break the loop.
Additionally I remember that while we introduced the module hierarchy
(#13009) we talked about splitting GHC.Driver.Types because it contained
various unrelated types and functions, but we never executed. I didn't
feel like making GHC.Driver.Types bigger with more unrelated Plugins
related types, so finally I bit the bullet and split GHC.Driver.Types.
As a consequence this patch moves a lot of things. I've tried to put
them into appropriate modules but nothing is set in stone.
Several other things moved to avoid loops.
* Removed Binary instances from GHC.Utils.Binary for random compiler
things
* Moved Typeable Binary instances into GHC.Utils.Binary.Typeable: they
import a lot of things that users of GHC.Utils.Binary don't want to
depend on.
* put everything related to Units/Modules under GHC.Unit:
GHC.Unit.Finder, GHC.Unit.Module.{ModGuts,ModIface,Deps,etc.}
* Created several modules under GHC.Types: GHC.Types.Fixity, SourceText,
etc.
* Split GHC.Utils.Error (into GHC.Types.Error)
* Finally removed GHC.Driver.Types
Note that this patch doesn't put loaded plugins into HscEnv. It's left
for another patch.
Bump haddock submodule
|
|
|
|
|
|
| |
pdocPrec was only used in GHC.Cmm.DebugBlock.pprUnwindExpr, so remove
it. OutputableP becomes a one-function class which might be better for
performance.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There are two signficant changes here:
* Ticket #18815 showed that we were missing some opportunities for
preInlineUnconditionally. The one-line fix is in the code for
GHC.Core.Opt.Simplify.Utils.preInlineUnconditionally, which now
switches off only for INLINE pragmas. I expanded
Note [Stable unfoldings and preInlineUnconditionally] to explain.
* When doing this I discovered a way in which preInlineUnconditionally
was occasionally /too/ eager. It's all explained in
Note [Occurrences in stable unfoldings] in GHC.Core.Opt.OccurAnal,
and the one-line change adding markAllMany to occAnalUnfolding.
I also got confused about what NoUserInline meant, so I've renamed
it to NoUserInlinePrag, and changed its pretty-printing slightly.
That led to soem error messate wibbling, and touches quite a few
files, but there is no change in functionality.
I did a nofib run. As expected, no significant changes.
Program Size Allocs
----------------------------------------
sphere -0.0% -0.4%
----------------------------------------
Min -0.0% -0.4%
Max -0.0% +0.0%
Geometric Mean -0.0% -0.0%
I'm allowing a max-residency increase for T10370, which seems
very irreproducible. (See comments on !4241.) There is always
sampling error for max-residency measurements; and in any case
the change shows up on some platforms but not others.
Metric Increase:
T10370
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Metric Decrease:
Naperian
T10421
T10421a
T10547
T12150
T12234
T12425
T13035
T18140
T18304
T5837
T6048
T13253-spj
T18282
T18223
T3064
T9961
Metric Increase
T13701
HFSKJH
|
| |
|
|
|
|
| |
A follow-up to !4020 (5830a12c46e7227c276a8a71213057595ee4fc04)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Haskell and Cmm parsers/lexers now report errors and warnings using ADTs
defined in GHC.Parser.Errors. They can be printed using functions in
GHC.Parser.Errors.Ppr.
Some of the errors provide hints with a separate ADT (e.g. to suggest to
turn on some extension). For now, however, hints are not consistent
across all messages. For example some errors contain the hints in the
main message. I didn't want to change any message with this patch. I
expect these changes to be discussed and implemented later.
Surprisingly, this patch enhances performance. On CI
(x86_64/deb9/hadrian, ghc/alloc):
parsing001 -11.5%
T13719 -2.7%
MultiLayerModules -3.5%
Naperian -3.1%
Bump haddock submodule
Metric Decrease:
MultiLayerModules
Naperian
T13719
parsing001
|
|
|
|
|
| |
There are still global variables but only 3 booleans instead of a single
DynFlags.
|
|
|
|
|
|
| |
Implements GHC Proposal #356
Updates the haddock submodule.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
* Don't depend on the selected backend to know if we print Asm or C
labels: we already have PprStyle to determine this. Moreover even when
a native backend is used (NCG, LLVM) we may want to C headers
containing pretty-printed labels, so it wasn't a good predicate
anyway.
* Make pretty-printing code clearer and avoid partiality
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Data.OldList exports a monomorphized singleton function but
it is not re-exported by Data.List. Adding the export to
Data.List causes a conflict with a 14-year old function of the
same name and type by SPJ in GHC.Utils.Misc. We can't just remove
this function because that leads to a problems when building
GHC with a stage0 compiler that does not have singleton in
Data.List yet. We also can't hide the function in GHC.Utils.Misc
since it is not possible to hide a function from a module if the
module does not export the function. To work around this, all
places where the Utils.Misc singleton was used now use a qualified
version like Utils.singleton and in GHC.Utils.Misc we are very
specific about which version we export.
|
| |
|
| |
|
|
|
|
|
| |
Add a type parameter for the environment required by OutputableP. It
avoids tying Platform with OutputableP.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Some types need a Platform value to be pretty-printed: CLabel, Cmm
types, instructions, etc.
Before this patch they had an Outputable instance and the Platform value
was obtained via sdocWithDynFlags. It meant that the *renderer* of the
SDoc was responsible of passing the appropriate Platform value (e.g. via
the DynFlags given to showSDoc). It put the burden of passing the
Platform value on the renderer while the generator of the SDoc knows the
Platform it is generating the SDoc for and there is no point passing a
different Platform at rendering time.
With this patch, we introduce a new OutputableP class:
class OutputableP a where
pdoc :: Platform -> a -> SDoc
With this class we still have some polymorphism as we have with `ppr`
(i.e. we can use `pdoc` on a variety of types instead of having a
dedicated `pprXXX` function for each XXX type).
One step closer removing `sdocWithDynFlags` (#10143) and supporting
several platforms (#14335).
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
source syntax (#18565)
Previously, we desugared and coverage checked plain guard trees as
described in Lower Your Guards. That caused (in !3849) quite a bit of
pain when we need to partially recover tree structure of the input
syntax to return covered sets for long-distance information, for
example.
In this refactor, I introduced a guard tree variant for each relevant
source syntax component of a pattern-match (mainly match groups, match,
GRHS, empty case, pattern binding). I made sure to share as much
coverage checking code as possible, so that the syntax-specific checking
functions are just wrappers around the more substantial checking
functions for the LYG primitives (`checkSequence`, `checkGrds`).
The refactoring payed off in clearer code and elimination of all panics
related to assumed guard tree structure and thus fixes #18565.
I also took the liberty to rename and re-arrange the order of functions
and comments in the module, deleted some dead and irrelevant Notes,
wrote some new ones and gave an overview module haddock.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
FastStrings can be compared in 2 ways: by Unique or lexically. We don't
want to bless one particular way with an "Ord" instance because it leads
to bugs (#18562) or to suboptimal code (e.g. using lexical comparison
while a Unique comparison would suffice).
UTF-8 encoding has the advantage that sorting strings by their encoded
bytes also sorts them by their Unicode code points, without having to
decode the actual code points. BUT GHC uses Modified UTF-8 which
diverges from UTF-8 by encoding \0 as 0xC080 instead of 0x00 (to avoid
null bytes in the middle of a String so that the string can still be
null-terminated). This patch adds a new `utf8CompareShortByteString`
function that performs sorting by bytes but that also takes Modified
UTF-8 into account. It is much more performant than decoding the strings
into [Char] to perform comparisons (which we did in the previous patch).
Bump haddock submodule
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When we pretty-print a UnitId for the user, we try to map it back to its
origin package name, version and component to print
"package-version:component" instead of some hash.
The UnitId type doesn't carry these information, so we have to look into
a UnitState to find them. This is why the Outputable instance of
UnitId used `sdocWithDynFlags` in order to access the `unitState` field
of DynFlags.
This is wrong for several reasons:
1. The DynFlags are accessed when the message is printed, not when it is
generated. So we could imagine that the unitState may have changed
in-between. Especially if we want to allow unit unloading.
2. We want GHC to support several independent sessions at once, hence
several UnitState. The current approach supposes there is a unique
UnitState as a UnitId doesn't indicate which UnitState to use.
See the Note [Pretty-printing UnitId] in GHC.Unit for the new approach
implemented by this patch.
One step closer to remove `sdocDynFlags` field from `SDocContext`
(#10143).
Fix #18124.
Also fix some Backpack code to use SDoc instead of String.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
With validate-x86_64-linux-deb9-hadrian:
T1969 -3.4% (threshold: +/-1%)
T3294 -3.3% (threshold: +/-1%)
T12707 -1.4% (threshold: +/-1%)
Additionally with validate-x86_64-linux-deb9-unreg-hadrian:
T4801 -2.4% (threshold: +/-2%)
T13035 -1.4% (threshold: +/-1%)
T13379 -2.4% (threshold: +/-2%)
ManyAlternatives -2.5% (threshold: +/-2%)
ManyConstructors -3.0% (threshold: +/-2%)
Metric Decrease:
T12707
T1969
T3294
ManyAlternatives
ManyConstructors
T13035
T13379
T4801
|
|
|
|
|
| |
This adds a new printing flag `sdocPrintTypeAbbreviations` that is used
specifically to avoid ghci printing 'type (->) = (->)' and 'type Type = Type'.
|
|
|
|
|
| |
The previous comment implies `nTimes n f` is either `f^{n+1}` or
`f^{2^n}` (when in fact it's `f^n`).
|
|
|
|
|
|
| |
* remove references to DynFlags in GHC.CmmToAsm.Dwarf
* add specific Dwarf options in NCGConfig instead of directly querying
the debug level
|
|
|
|
|
|
| |
The note has been rewritten by @simonpj in !3851
[skip ci]
|
|
|
|
|
|
|
|
|
| |
- put panic related functions into GHC.Utils.Panic
- put trace related functions using DynFlags in GHC.Driver.Ppr
One step closer making Outputable fully independent of DynFlags.
Bump haddock submodule
|
|
|
|
|
|
|
|
|
| |
Also fix its slightly wrong comment
Metric Decrease:
T5030
T12227
T12545
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As suspected by @simonpj in #18535, avoiding allocations in
`GHC.Utils.Misc.splitAtList` when there are no leftover arguments is
beneficial for performance:
On CI validate-x86_64-linux-deb9-hadrian:
T12227 -7%
T12545 -12.3%
T5030 -10%
T9872a -2%
T9872b -2.1%
T9872c -2.5%
Metric Decrease:
T12227
T12545
T5030
T9872a
T9872b
T9872c
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
| |
Currently we're passing a indexWord8OffAddr# type function to
utf8DecodeLazy# which then passes it on to utf8DecodeChar#. By passing one
of utf8DecodeCharAddr# or utf8DecodeCharByteArray# instead we benefit from
the inlining and specialization already done for those.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There are multiple reasons we want this:
- Fewer allocations: ByteString has 3 fields, ShortByteString just has one.
- ByteString memory is pinned:
- This can cause fragmentation issues (see for example #13110) but also
- makes using FastStrings in compact regions impossible.
Metric Decrease:
T5837
T12150
T12234
T12425
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Haddock comments are, first and foremost, comments. It's very annoying
to incorporate them into the grammar. We can take advantage of an
important property: adding a Haddock comment does not change the parse
tree in any way other than wrapping some nodes in HsDocTy and the like
(and if it does, that's a bug).
This patch implements the following:
* Accumulate Haddock comments with their locations in the P monad.
This is handled in the lexer.
* After parsing, do a pass over the AST to associate Haddock comments
with AST nodes using location info.
* Report the leftover comments to the user as a warning (-Winvalid-haddock).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Executing on the plan described in #17582, this patch changes the way if expressions
are handled in the compiler in the presence of rebindable syntax. We get rid of the
SyntaxExpr field of HsIf and instead, when rebindable syntax is on, we rewrite the HsIf
node to the appropriate sequence of applications of the local `ifThenElse` function.
In order to be able to report good error messages, with expressions as they were
written by the user (and not as desugared by the renamer), we make use of TTG
extensions to extend GhcRn expression ASTs with an `HsExpansion` construct, which
keeps track of a source (GhcPs) expression and the desugared (GhcRn) expression that
it gives rise to. This way, we can typecheck the latter while reporting the former in
error messages.
In order to discard the error context lines that arise from typechecking the desugared
expressions (because they talk about expressions that the user has not written), we
carefully give a special treatment to the nodes fabricated by this new renaming-time
transformation when typechecking them. See Note [Rebindable syntax and HsExpansion]
for more details. The note also includes a recipe to apply the same treatment to
other rebindable constructs.
Tests 'rebindable11' and 'rebindable12' have been added to make sure we report
identical error messages as before this patch under various circumstances.
We also now disable rebindable syntax when processing untyped TH quotes, as per
the discussion in #18102 and document the interaction of rebindable syntax and
Template Haskell, both in Note [Template Haskell quotes and Rebindable Syntax]
and in the user guide, adding a test to make sure that we do not regress in
that regard.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This fixes #17667 and should help to avoid such issues going forward.
The changes are mostly mechanical in nature. With two notable
exceptions.
* The register allocator.
The register allocator references registers by distinct uniques.
However they come from the types of VirtualReg, Reg or Unique in
various places. As a result we sometimes cast the key type of the
map and use functions which operate on the now typed map but take
a raw Unique as actual key. The logic itself has not changed it
just becomes obvious where we do so now.
* <Type>Env Modules.
As an example a ClassEnv is currently queried using the types `Class`,
`Name`, and `TyCon`. This is safe since for a distinct class value all
these expressions give the same unique.
getUnique cls
getUnique (classTyCon cls)
getUnique (className cls)
getUnique (tcName $ classTyCon cls)
This is for the most part contained within the modules defining the
interface. However it requires us to play dirty when we are given a
`Name` to lookup in a `UniqFM Class a` map. But again the logic did
not change and it's for the most part hidden behind the Env Module.
Some of these cases could be avoided by refactoring but this is left
for future work.
We also bump the haddock submodule as it uses UniqFM.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This updates haddock comments only.
This patch focuses to update for hyperlinks in GHC API's haddock comments,
because broken links especially discourage newcomers.
This includes the following hierarchies:
- GHC.Iface.*
- GHC.Llvm.*
- GHC.Rename.*
- GHC.Tc.*
- GHC.HsToCore.*
- GHC.StgToCmm.*
- GHC.CmmToAsm.*
- GHC.Runtime.*
- GHC.Unit.*
- GHC.Utils.*
- GHC.SysTools.*
|
|
|
|
| |
It avoids having to use DynFlags to reach for pprUserLength.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
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.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
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
|