| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
| |
Progress towards #17957
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We were considering all Typeable evidence to be "BuiltinInstance"s which
meant the stage restriction was going unchecked. In-fact, typeable has
evidence and so we need to apply the stage restriction.
This is
complicated by the fact we don't generate typeable evidence and the
corresponding DFunIds until after typechecking is concluded so we
introcue a new `InstanceWhat` constructor, BuiltinTypeableInstance which
records whether the evidence is going to be local or not.
Fixes #21547
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
With this change, `Backend` becomes an abstract type
(there are no more exposed value constructors).
Decisions that were formerly made by asking "is the
current back end equal to (or different from) this named value
constructor?" are now made by interrogating the back end about
its properties, which are functions exported by `GHC.Driver.Backend`.
There is a description of how to migrate code using `Backend` in the
user guide.
Clients using the GHC API can find a backdoor to access the Backend
datatype in GHC.Driver.Backend.Internal.
Bumps haddock submodule.
Fixes #20927
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This firstly caused spurious output to be emitted (as evidenced by
#21555) but even worse caused a massive coercion to be attempted to be
printed (> 200k terms) which would invariably eats up all the memory of
your computer.
The good news is that removing this trace allows the program to compile
to completion, the bad news is that the program exhibits a core lint
error (on 9.0.2) but not any other releases it seems.
Fixes #21577 and #21555
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit adds module `GHC.Cmm.Dominators`, which provides a wrapper
around two existing algorithms in GHC: the Lengauer-Tarjan dominator
analysis from the X86 back end and the reverse postorder ordering from
the Cmm Dataflow framework. Issue #20726 proposes that we evaluate
some alternatives for dominator analysis, but for the time being, the
best path forward is simply to use the existing analysis on
`CmmGraph`s.
This commit addresses a bullet in #21200.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
LlvmConfig contains information read from llvm-passes and llvm-targets
files in GHC's top directory. Reading these files is done only when
needed (i.e. when the LLVM backend is used) and cached for the whole
compiler session. This patch changes the way this is done:
- Split LlvmConfig into LlvmConfig and LlvmConfigCache
- Store LlvmConfigCache in HscEnv instead of DynFlags: there is no
good reason to store it in DynFlags. As it is fixed per session, we
store it in the session state instead (HscEnv).
- Initializing LlvmConfigCache required some changes to driver functions
such as newHscEnv. I've used the opportunity to untangle initHscEnv
from initGhcMonad (in top-level GHC module) and to move it to
GHC.Driver.Main, close to newHscEnv.
- I've also made `cmmPipeline` independent of HscEnv in order to remove
the call to newHscEnv in regalloc_unit_tests.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When generating an SRT for a recursive group, GHC.Cmm.Info.Build.oneSRT
filters out recursive references, as described in Note [recursive SRTs].
However, doing so for static functions would be unsound, for the reason
described in Note [Invalid optimisation: shortcutting].
However, the same argument applies to static data constructor
applications, as we discovered in #20959. Fix this by ensuring that
static data constructor applications are included in recursive SRTs.
The approach here is not entirely satisfactory, but it is a starting
point.
Fixes #20959.
|
|
|
|
|
|
|
|
|
|
| |
Here we implement a few measures to improve the clarity of the CAF
analysis implementation. Specifically:
* Use CafInfo instead of Bool since the former is more descriptive
* Rename CAFLabel to CAFfyLabel, since not all CAFfyLabels are in fact
CAFs
* Add numerous comments
|
|
|
|
|
|
|
|
|
|
|
|
| |
This MR fixes a bad bug, where the withDict was inlined too
vigorously, which in turn made the type-class Specialiser generate
a bogus specialisation, because it saw the same overloaded function
applied to two /different/ dictionaries.
Solution: inline `withDict` later. See (WD8) of Note [withDict]
in GHC.HsToCore.Expr
See #21575, which is fixed by this change.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We were using defaultSDocContext for pprTrace, which suppresses
lots of useful infomation. This small MR adds
GHC.Utils.Outputable.traceSDocContext
and uses it for pprTrace and pprTraceUserWarning.
traceSDocContext is a global, and hence not influenced by flags,
but that seems unavoidable. But I made the sdocPprDebug bit
controlled by unsafeHasPprDebug, since we have the latter for
exactly this purpose.
Fixes #21569
|
|
|
|
| |
for better disambiguation (#17420)
|
|
|
|
|
|
|
|
|
|
| |
Here we introduce proper support for compilation of C++ objects. This
includes:
* logic in `configure` to detect the C++ toolchain and propagating this
information into the `settings` file
* logic in the driver to use the C++ toolchain when compiling C++
sources
|
| |
|
|
|
|
|
|
|
|
| |
- Remove groupWithName (unused)
- Use the RuntimeRepType synonym where possible
- Replace getUniqueM + mkSysLocalOrCoVar with mkSysLocalOrCoVarM
No functional changes.
|
|
|
|
|
| |
The lack of INLNE arity was exposed by #21531. The fix is
simple enough, if a bit clumsy.
|
| |
|
|
|
|
|
|
|
|
|
| |
Change mulArrow to allow for printing of correct application precedence
where necessary and update callers of mulArrow to reflect this.
As part of this, move mulArrow from GHC/Utils/Outputtable to GHC/Iface/Type.
Fixes #20315
|
|
|
|
|
| |
Metric Decrease:
T16875
|
| |
|
| |
|
| |
|
|
|
|
| |
This missing guard gave rise to #21519.
|
|
|
|
|
|
|
|
|
|
|
| |
Ticket #21489 showed that the saturation mechanism for
DFuns (see Note Specialising DFuns) should use both
UnspecType and UnspecArg.
We weren't doing that; but this MR fixes that problem.
No test case because it's hard to tickle, but it showed up in
Gergo's work with GHC-as-a-library.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The code
do; a <- doAsync; b
Generated an incorrect Anchor for the statement list that starts after
the first semicolon.
This commit fixes it.
Closes #20256
|
|
|
|
|
|
| |
The old code used by via-C backend didn't handle the sign bit of NaN.
See #21043.
|
|
|
|
|
|
|
| |
We assumed the wrapper for an unlifted binding is the identity,
but as #21516 showed, that is no always true.
Solution is simple: use it.
|
|
|
|
|
|
|
|
|
|
| |
This fixes #21479
See Note [Unquantified tyvars in a pattern synonym]
While doing this, I found that some error messages pointed at the
pattern synonym /name/, rather than the /declaration/ so I widened the
SrcSpan to encompass the declaration.
|
|
|
|
|
|
|
|
|
|
|
|
| |
It seems like it was just an oversight to use the incorrect DynFlags
(global rather than local) when implementing these two options. Using
the local flags allows users to request these intermediate files get
cleaned up, which works fine in --make mode because
1. Interface files are stored in memory
2. Object files are only cleaned at the end of session (after link)
Fixes #21349
|
|
|
|
|
|
| |
Add a paragraph that clarifies that `occurAnalysePgm` finding out-of-order
references, and thus needing to glom, is not a cause for concern when its
root cause is rewrite rules.
|
| |
|
|
|
|
| |
mapAdjust is more efficient than mapAlter.
|
|
|
|
|
|
|
| |
Fixes #21362
Metric Decrease:
T16875
|
| |
|
|
|
|
|
|
|
|
|
|
| |
There is a mis-match between the TH representation of OPAQUE pragmas and
GHC's internal representation due to how OPAQUE pragmas disallow phase
annotations. It seemed most in keeping to just fix the wired in name
issue by adding a special case to the desugaring of INLINE pragmas
rather than making TH/GHC agree with how the representation should look.
Fixes #21463
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The following is currently rejected:
```haskell
-- F is an Applicative but not a Monad
x :: F (Int, Int)
x = do
a <- pure 0
let b = 1
pure (a, b)
```
This has bitten me multiple times. This MR contains a simple fix:
only allow a "let only" segment to be merged with the next (and not
the previous) segment. As a result, when the last one or more
statements before pure/return are `LetStmt`s, there will be one
more segment containing only those `LetStmt`s.
Note that if the `let` statement mentions a name bound previously, then
the program is still rejected, for example
```haskell
x = do
a <- pure 0
let b = a + 1
pure (a, b)
```
or the example in #18559. To support this would require a more
complex approach, but this is IME much less common than the
previous case.
|
|
|
|
|
|
|
|
| |
Previously we only preserved the bottom 64-bits of the callee-saved
128-bit XMM registers, in violation of the Win64 calling convention.
Fix this.
Fixes #21465.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
applications
The main fix is that in addVoidWorkerArg we now add the argument to the front.
This fixes #21448.
-------------------------
Metric Decrease:
T16875
-------------------------
|
|
|
|
|
|
|
|
|
|
|
| |
IntMap.size is O(n). The new code should be slightly more efficient.
The transformation of GHC.CmmToAsm.CFG.calcFreqs.nodeCount can be
described formally as the transformation:
(\sum_{0}^{n-1} \sum_{0}^{k-1} i_nk) + n
==>
(\sum_{0}^{n-1} 1 + \sum_{0}^{k-1} i_nk)
|
| |
|
|
|
|
|
|
|
|
| |
The previous patch forgot to account for a type such as
Any @(TYPE (BoxedRep l))
for a quantified levity variable l.
|
|
|
|
|
|
| |
Our error reporting in generated code (via desugaring before
typechecking) only worked when the generated code was just a simple
call. This commit makes it work in nested cases.
|
|
|
|
|
|
|
|
|
|
|
| |
The LaTeX documentation generator does not seem to have been used for
quite some time, so the LaTeX-to-Haddock preprocessing step has become a
pointless complication that makes documenting the contents of GHC.Prim
needlessly difficult. This commit replaces the LaTeX syntax with the
Haddock it would have been converted into, anyway, though with an
additional distinction: it uses single quotes in places to instruct
Haddock to generate hyperlinks to bindings. This improves the quality of
the generated output.
|
|
|
|
|
|
|
|
|
|
|
| |
For expressions like `(scc<cc_name> primOp#) arg1` we should also look
at arg1 to determine if we call primOp# at a fixed runtime rep.
This is what corePrep already does but CoreLint didn't yet. This patch
will bring them in sync in this regard.
It also uses tickishFloatable in CorePrep instead of CorePrep having
it's own slightly differing definition of when a tick is floatable.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
See the new `Note [SubDemand denotes at least one evaluation]`.
A demand `n :* sd` on a let binder `x=e` now means
> "`x` was evaluated `n` times and in any program trace it is evaluated, `e` is
> evaluated deeply in sub-demand `sd`."
The "any time it is evaluated" premise is what this patch adds. As a result,
we get better nested strictness. For example (T21081)
```hs
f :: (Bool, Bool) -> (Bool, Bool)
f pr = (case pr of (a,b) -> a /= b, True)
-- before: <MP(L,L)>
-- after: <MP(SL,SL)>
g :: Int -> (Bool, Bool)
g x = let y = let z = odd x in (z,z) in f y
```
The change in demand signature "before" to "after" allows us to case-bind `z`
here.
Similarly good things happen for the `sd` in call sub-demands `Cn(sd)`, which
allows for more eta-reduction (which is only sound with `-fno-pedantic-bottoms`,
albeit).
We also fix #21085, a surprising inconsistency with `Poly` to `Call` sub-demand
expansion.
In an attempt to fix a regression caused by less inlining due to eta-reduction
in T15426, I eta-expanded the definition of `elemIndex` and `elemIndices`, thus
fixing #21345 on the go.
The main point of this patch is that it fixes #21081 and #21133.
Annoyingly, I discovered that more precise demand signatures for join points can
transform a program into a lazier program if that join point gets floated to the
top-level, see #21392. There is no simple fix at the moment, but !5349 might.
Thus, we accept a ~5% regression in `MultiLayerModulesTH_OneShot`, where #21392
bites us in `addListToUniqDSet`. T21392 reliably reproduces the issue.
Surprisingly, ghc/alloc perf on Windows improves much more than on other jobs, by
0.4% in the geometric mean and by 2% in T16875.
Metric Increase:
MultiLayerModulesTH_OneShot
Metric Decrease:
T16875
|
| |
|
|
|
|
|
| |
Note [Nullary unboxed tuple] was removed in e9e61f18a548b70693f4.
This codepath is tested by T15696_3.
|
|
|
|
|
|
|
|
|
|
|
|
| |
bottoming.
We used to check the divergence and that the number of arguments > arity.
But arity zero represents unknown arity so this was subtly broken for a long time!
We would check if the saturated function diverges, and if we applied >=arity arguments.
But for unknown arity functions any number of arguments is >=idArity.
This fixes #21440.
|