| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The big payload of this patch is:
Add an AnonArgFlag to the FunTy constructor
of Type, so that
(FunTy VisArg t1 t2) means (t1 -> t2)
(FunTy InvisArg t1 t2) means (t1 => t2)
The big payoff is that we have a simple, local test to make
when decomposing a type, leading to many fewer calls to
isPredTy. To me the code seems a lot tidier, and probably
more efficient (isPredTy has to take the kind of the type).
See Note [Function types] in TyCoRep.
There are lots of consequences
* I made FunTy into a record, so that it'll be easier
when we add a linearity field, something that is coming
down the road.
* Lots of code gets touched in a routine way, simply because it
pattern matches on FunTy.
* I wanted to make a pattern synonym for (FunTy2 arg res), which
picks out just the argument and result type from the record. But
alas the pattern-match overlap checker has a heart attack, and
either reports false positives, or takes too long. In the end
I gave up on pattern synonyms.
There's some commented-out code in TyCoRep that shows what I
wanted to do.
* Much more clarity about predicate types, constraint types
and (in particular) equality constraints in kinds. See TyCoRep
Note [Types for coercions, predicates, and evidence]
and Note [Constraints in kinds].
This made me realise that we need an AnonArgFlag on
AnonTCB in a TyConBinder, something that was really plain
wrong before. See TyCon Note [AnonTCB InivsArg]
* When building function types we must know whether we
need VisArg (mkVisFunTy) or InvisArg (mkInvisFunTy).
This turned out to be pretty easy in practice.
* Pretty-printing of types, esp in IfaceType, gets
tidier, because we were already recording the (->)
vs (=>) distinction in an ad-hoc way. Death to
IfaceFunTy.
* mkLamType needs to keep track of whether it is building
(t1 -> t2) or (t1 => t2). See Type
Note [mkLamType: dictionary arguments]
Other minor stuff
* Some tidy-up in validity checking involving constraints;
Trac #16263
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch makes the following improvement:
- Automatically records test metrics (per test environment) so that
the programmer need not supply nor update expected values in *.T
files.
- On expected metric changes, the programmer need only indicate the
direction of change in the git commit message.
- Provides a simple python tool "perf_notes.py" to compare metrics
over time.
Issues:
- Using just the previous commit allows performance to drift with each
commit.
- Currently we allow drift as we have a preference for minimizing
false positives.
- Some possible alternatives include:
- Use metrics from a fixed commit per test: the last commit that
allowed a change in performance (else the oldest metric)
- Or use some sort of aggregate since the last commit that allowed
a change in performance (else all available metrics)
- These alternatives may result in a performance issue (with the
test driver) having to heavily search git commits/notes.
- Run locally, performance tests will trivially pass unless the tests
were run locally on the previous commit. This is often not the case
e.g. after pulling recent changes.
Previously, *.T files contain statements such as:
```
stats_num_field('peak_megabytes_allocated', (2, 1))
compiler_stats_num_field('bytes allocated',
[(wordsize(64), 165890392, 10)])
```
This required the programmer to give the expected values and a tolerance
deviation (percentage). With this patch, the above statements are
replaced with:
```
collect_stats('peak_megabytes_allocated', 5)
collect_compiler_stats('bytes allocated', 10)
```
So that programmer must only enter which metrics to test and a tolerance
deviation. No expected value is required. CircleCI will then run the
tests per test environment and record the metrics to a git note for that
commit and push them to the git.haskell.org ghc repo. Metrics will be
compared to the previous commit. If they are different by the tolerance
deviation from the *.T file, then the corresponding test will fail. By
adding to the git commit message e.g.
```
# Metric (In|De)crease <metric(s)> <options>: <tests>
Metric Increase ['bytes allocated', 'peak_megabytes_allocated'] \
(test_env='linux_x86', way='default'):
Test012, Test345
Metric Decrease 'bytes allocated':
Test678
Metric Increase:
Test711
```
This will allow the noted changes (letting the test pass). Note that by
omitting metrics or options, the change will apply to all possible
metrics/options (i.e. in the above, an increase for all metrics in all
test environments is allowed for Test711)
phabricator will use the message in the description
Reviewers: bgamari, hvr
Reviewed By: bgamari
Subscribers: rwbarton, carter
GHC Trac Issues: #12758
Differential Revision: https://phabricator.haskell.org/D5059
|
|
|
|
|
|
|
| |
This completes the 2nd phase of the Semigroup=>Monoid Proposal (SMP)
initiated in 8ae263ceb3566a7c82336400b09cb8f381217405.
This updates a couple submodules to address <> naming clashes.
|
|
|
|
|
|
|
|
|
|
| |
Test Plan: Validate
Reviewers: austin
Subscribers: rwbarton, thomie
Differential Revision: https://phabricator.haskell.org/D3303
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The fundamental problem with `type UniqSet = UniqFM` is that `UniqSet`
has a key invariant `UniqFM` does not. For example, `fmap` over
`UniqSet` will generally produce nonsense.
* Upgrade `UniqSet` from a type synonym to a newtype.
* Remove unused and shady `extendVarSet_C` and `addOneToUniqSet_C`.
* Use cached unique in `tyConsOfType` by replacing
`unitNameEnv (tyConName tc) tc` with `unitUniqSet tc`.
Reviewers: austin, hvr, goldfire, simonmar, niteria, bgamari
Reviewed By: niteria
Subscribers: thomie
Differential Revision: https://phabricator.haskell.org/D3146
|
|
|
|
|
|
|
| |
The desugarer was producing an empty Rec group, which is never
supposed to happen. This small patch stops that happening.
Next up: Lint should check.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Originally, everything that is not in WHNF (`exprIsWHNF`) is considered
a thunk, not eta-expanded, to avoid losing any sharing. This is also how
the published papers on Call Arity describe it.
In practice, there are thunks that do a just little work, such as
pattern-matching on a variable, and the benefits of eta-expansion likely
oughtweigh the cost of doing that repeatedly. Therefore, this
implementation of Call Arity considers everything that is not cheap
(`exprIsCheap`) as a thunk.
Nofib reports -2.58% allocations for scs and -40.93% allocation for
wheel-sieve1; the latter has - 2.92% runtime.
|
|
|
|
|
|
|
|
| |
This eradicates varSetElems from the codebase. This function
used to introduce nondeterminism.
I've also documented benign nondeterminism in three places.
GHC Trac: #4012
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This improves how the Call Arity deals with "boring" variables. Boring
variables are those where it does not bother to include in the analysis
result, so whenever something is looked up in the analysis result, we
have to make a conservative assumption about them.
Previously, we extended the result with such conservative information
about them, to keep the code uniform, but that could blow up the amount
of data passed around, even if only temporarily, and slowed things down.
We now pass around an explicit list (well, set) of variable that are
boring and take that into account whenever we use the result. Not as
pretty, but noticably faster.
|
|
|
|
| |
Signed-off-by: Austin Seipp <austin@well-typed.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
This is a draft of the patch that is sent for review.
In this patch required changes in linter were introduced
and actual check:
- new helper function: primRepSizeB
- primRep check for floating
- Add access to dynamic flags in linter.
- Implement additional lint rules.
Reviewers: austin, goldfire, simonpj
Reviewed By: simonpj
Subscribers: thomie
Differential Revision: https://phabricator.haskell.org/D637
GHC Trac Issues: #9122
|
|
|
|
|
| |
This reverts commit 9711f78f790d10d914e08851544c6fc96f9a030a, as it's
causing build phailures in phabricator.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Summary:
* Adjusts performance tests
* Change ghcpkg05.stderr-mingw32 to match ghcpkg05.stderr
Test Plan: Ran 'sh validate' and observed fewer test failures afterwards
Reviewers: austin
Reviewed By: austin
Differential Revision: https://phabricator.haskell.org/D191
|
|
|
|
|
|
|
|
|
|
|
|
| |
These did not yet trigger a failure, but are more than 1% away from the
expected value. Since I now start collecting logs to investigate
deviations from the expected value, it makes sense to reset them. This
way we know that every significat deviation was caused since this
commit.
I only updated bytes_allocated numbers, as these are (mostly)
deterministic. Other depend, AFAIK, on sampling timing, so I did not
bother.
|
|
|
|
|
|
|
|
|
| |
It's a bit confusing to have .gitignore files spread all over the
filesystem. This commit tries to consolidate those into one .gitignore
file per component. Moreover, we try to describe files to be ignored which
happen to have a common identifying pattern by glob patterns.
Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>
|
|
|
|
|
|
|
|
| |
I used this shell command to automatically generate the lists:
for i in `git ls-files -o --exclude-standard --directory`; do echo "`basename $i`" >> "`dirname "$i"`/.gitignore"; done
Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>
|
|
|
|
| |
Many are improvements!
|
| |
|
|
|
|
|
|
|
|
| |
Even if the recursion is a nice tail-call only recusion, we'd stil be
calling the thunk multiple times and eta-expansion would be wrong.
Includes a [Note].
(Also shows the disadvantage of unit tests: They had the same bug.)
|
|
|
|
| |
(Otherwise the analysis was wrong, as covered by the new test case.)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch improves the call arity analysis in various ways.
Most importantly, it enriches the analysis result information so that
when looking at a call, we do not have to make a random choice about
what side we want to take the information from. Instead we can combine
the results in a way that does not lose valuable information.
To do so, besides the incoming arities, we store remember "what can be
called with what", i.e. an undirected graph between the (interesting)
free variables of an expression. Of course it makes combining the
results a bit more tricky (especially mutual recursion), but still
doable.
The actually implemation of the graph structure is abstractly put away
in a module of its own (UnVarGraph.hs)
The implementation is geared towards efficiently representing the graphs
that we need (which can contain large complete and large complete
bipartite graphs, which would be huge in other representations). If
someone feels like designing data structures: There is surely some
speed-up to be obtained by improving that data structure.
Additionally, the analysis now takes into account that if a RHS stays a
thunk, then its calls happen only once, even if the variables the RHS is
bound to is evaluated multiple times, or is part of a recursive group.
|
|
|
|
| |
nice numbers coming from these micro-benchmarks.
|
| |
|
| |
|
|
|
|
| |
by elaborating the domain a bit.
|
|
This also sets precedence for testing internals of GHC directly, i.e.
without trying to come up with Haskell code and observable effects.
Let's see how that goes.
I put all the tests (including those where the analysis could do better)
in one file because starting the GHC API is quite slow.
|