summaryrefslogtreecommitdiff
path: root/testsuite/tests/simplStg
Commit message (Collapse)AuthorAgeFilesLines
* Misc cleanupKrzysztof Gogolewski2023-01-111-1/+1
| | | | | | | | - Remove unused mkWildEvBinder - Use typeTypeOrConstraint - more symmetric and asserts that that the type is Type or Constraint - Fix escape sequences in Python; they raise a deprecation warning with -Wdefault
* Fix GHCis interaction with tag inference.Andreas Klebinger2022-10-185-0/+27
| | | | | | | | | | | | | | | | | | | | | | | | | I had assumed that wrappers were not inlined in interactive mode. Meaning we would always execute the compiled wrapper which properly takes care of upholding the strict field invariant. This turned out to be wrong. So instead we now run tag inference even when we generate bytecode. In that case only for correctness not performance reasons although it will be still beneficial for runtime in some cases. I further fixed a bug where GHCi didn't tag nullary constructors properly when used as arguments. Which caused segfaults when calling into compiled functions which expect the strict field invariant to be upheld. Fixes #22042 and #21083 ------------------------- Metric Increase: T4801 Metric Decrease: T13035 -------------------------
* Don't use isUnliftedType in isTaggedsheaf2022-09-212-0/+47
| | | | | | | | | The function GHC.Stg.InferTags.Rewrite.isTagged can be given the Id of a join point, which might be representation polymorphic. This would cause the call to isUnliftedType to crash. It's better to use typeLevity_maybe instead. Fixes #22212
* Tag inference: Fix #21954 by retaining tagsigs of vars in function position.Andreas Klebinger2022-09-153-0/+179
| | | | | | | | | | For an expression like: case x of y Con z -> z If we also retain the tag sig for z we can generate code to immediately return it rather than calling out to stg_ap_0_fast.
* STG: only print cost-center if asked toSylvain Henry2022-05-091-2/+2
|
* Tag inference work.Andreas Klebinger2022-02-124-1/+26
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This does three major things: * Enforce the invariant that all strict fields must contain tagged pointers. * Try to predict the tag on bindings in order to omit tag checks. * Allows functions to pass arguments unlifted (call-by-value). The former is "simply" achieved by wrapping any constructor allocations with a case which will evaluate the respective strict bindings. The prediction is done by a new data flow analysis based on the STG representation of a program. This also helps us to avoid generating redudant cases for the above invariant. StrictWorkers are created by W/W directly and SpecConstr indirectly. See the Note [Strict Worker Ids] Other minor changes: * Add StgUtil module containing a few functions needed by, but not specific to the tag analysis. ------------------------- Metric Decrease: T12545 T18698b T18140 T18923 LargeRecord Metric Increase: LargeRecord ManyAlternatives ManyConstructors T10421 T12425 T12707 T13035 T13056 T13253 T13253-spj T13379 T15164 T18282 T18304 T18698a T1969 T20049 T3294 T4801 T5321FD T5321Fun T783 T9233 T9675 T9961 T19695 WWRec -------------------------
* Nested CPR light unleashed (#18174)Sebastian Graf2021-09-303-16/+11
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This patch enables worker/wrapper for nested constructed products, as described in `Note [Nested CPR]`. The machinery for expressing Nested CPR was already there, since !5054. Worker/wrapper is equipped to exploit Nested CPR annotations since !5338. CPR analysis already handles applications in batches since !5753. This patch just needs to flip a few more switches: 1. In `cprTransformDataConWork`, we need to look at the field expressions and their `CprType`s to see whether the evaluation of the expressions terminates quickly (= is in HNF) or if they are put in strict fields. If that is the case, then we retain their CPR info and may unbox nestedly later on. More details in `Note [Nested CPR]`. 2. Enable nested `ConCPR` signatures in `GHC.Types.Cpr`. 3. In the `asConCpr` call in `GHC.Core.Opt.WorkWrap.Utils`, pass CPR info of fields to the `Unbox`. 4. Instead of giving CPR signatures to DataCon workers and wrappers, we now have `cprTransformDataConWork` for workers and treat wrappers by analysing their unfolding. As a result, the code from GHC.Types.Id.Make went away completely. 5. I deactivated worker/wrappering for recursive DataCons and wrote a function `isRecDataCon` to detect them. We really don't want to give `repeat` or `replicate` the Nested CPR property. See Note [CPR for recursive data structures] for which kind of recursive DataCons we target. 6. Fix a couple of tests and their outputs. I also documented that CPR can destroy sharing and lead to asymptotic increase in allocations (which is tracked by #13331/#19326) in `Note [CPR for data structures can destroy sharing]`. Nofib results: ``` -------------------------------------------------------------------------------- Program Allocs Instrs -------------------------------------------------------------------------------- ben-raytrace -3.1% -0.4% binary-trees +0.8% -2.9% digits-of-e2 +5.8% +1.2% event +0.8% -2.1% fannkuch-redux +0.0% -1.4% fish 0.0% -1.5% gamteb -1.4% -0.3% mkhprog +1.4% +0.8% multiplier +0.0% -1.9% pic -0.6% -0.1% reptile -20.9% -17.8% wave4main +4.8% +0.4% x2n1 -100.0% -7.6% -------------------------------------------------------------------------------- Min -95.0% -17.8% Max +5.8% +1.2% Geometric Mean -2.9% -0.4% ``` The huge wins in x2n1 (loopy list) and reptile (see #19970) are due to refraining from unboxing (:). Other benchmarks like digits-of-e2 or wave4main regress because of that. Ultimately there are no great improvements due to Nested CPR alone, but at least it's a win. Binary sizes decrease by 0.6%. There are a significant number of metric decreases. The most notable ones (>1%): ``` ManyAlternatives(normal) ghc/alloc 771656002.7 762187472.0 -1.2% ManyConstructors(normal) ghc/alloc 4191073418.7 4114369216.0 -1.8% MultiLayerModules(normal) ghc/alloc 3095678333.3 3128720704.0 +1.1% PmSeriesG(normal) ghc/alloc 50096429.3 51495664.0 +2.8% PmSeriesS(normal) ghc/alloc 63512989.3 64681600.0 +1.8% PmSeriesV(normal) ghc/alloc 62575424.0 63767208.0 +1.9% T10547(normal) ghc/alloc 29347469.3 29944240.0 +2.0% T11303b(normal) ghc/alloc 46018752.0 47367576.0 +2.9% T12150(optasm) ghc/alloc 81660890.7 82547696.0 +1.1% T12234(optasm) ghc/alloc 59451253.3 60357952.0 +1.5% T12545(normal) ghc/alloc 1705216250.7 1751278952.0 +2.7% T12707(normal) ghc/alloc 981000472.0 968489800.0 -1.3% GOOD T13056(optasm) ghc/alloc 389322664.0 372495160.0 -4.3% GOOD T13253(normal) ghc/alloc 337174229.3 341954576.0 +1.4% T13701(normal) ghc/alloc 2381455173.3 2439790328.0 +2.4% BAD T14052(ghci) ghc/alloc 2162530642.7 2139108784.0 -1.1% T14683(normal) ghc/alloc 3049744728.0 2977535064.0 -2.4% GOOD T14697(normal) ghc/alloc 362980213.3 369304512.0 +1.7% T15164(normal) ghc/alloc 1323102752.0 1307480600.0 -1.2% T15304(normal) ghc/alloc 1304607429.3 1291024568.0 -1.0% T16190(normal) ghc/alloc 281450410.7 284878048.0 +1.2% T16577(normal) ghc/alloc 7984960789.3 7811668768.0 -2.2% GOOD T17516(normal) ghc/alloc 1171051192.0 1153649664.0 -1.5% T17836(normal) ghc/alloc 1115569746.7 1098197592.0 -1.6% T17836b(normal) ghc/alloc 54322597.3 55518216.0 +2.2% T17977(normal) ghc/alloc 47071754.7 48403408.0 +2.8% T17977b(normal) ghc/alloc 42579133.3 43977392.0 +3.3% T18923(normal) ghc/alloc 71764237.3 72566240.0 +1.1% T1969(normal) ghc/alloc 784821002.7 773971776.0 -1.4% GOOD T3294(normal) ghc/alloc 1634913973.3 1614323584.0 -1.3% GOOD T4801(normal) ghc/alloc 295619648.0 292776440.0 -1.0% T5321FD(normal) ghc/alloc 278827858.7 276067280.0 -1.0% T5631(normal) ghc/alloc 586618202.7 577579960.0 -1.5% T5642(normal) ghc/alloc 494923048.0 487927208.0 -1.4% T5837(normal) ghc/alloc 37758061.3 39261608.0 +4.0% T9020(optasm) ghc/alloc 257362077.3 254672416.0 -1.0% T9198(normal) ghc/alloc 49313365.3 50603936.0 +2.6% BAD T9233(normal) ghc/alloc 704944258.7 685692712.0 -2.7% GOOD T9630(normal) ghc/alloc 1476621560.0 1455192784.0 -1.5% T9675(optasm) ghc/alloc 443183173.3 433859696.0 -2.1% GOOD T9872a(normal) ghc/alloc 1720926653.3 1693190072.0 -1.6% GOOD T9872b(normal) ghc/alloc 2185618061.3 2162277568.0 -1.1% GOOD T9872c(normal) ghc/alloc 1765842405.3 1733618088.0 -1.8% GOOD TcPlugin_RewritePerf(normal) ghc/alloc 2388882730.7 2365504696.0 -1.0% WWRec(normal) ghc/alloc 607073186.7 597512216.0 -1.6% T9203(normal) run/alloc 107284064.0 102881832.0 -4.1% haddock.Cabal(normal) run/alloc 24025329589.3 23768382560.0 -1.1% haddock.base(normal) run/alloc 25660521653.3 25370321824.0 -1.1% haddock.compiler(normal) run/alloc 74064171706.7 73358712280.0 -1.0% ``` The biggest exception to the rule is T13701 which seems to fluctuate as usual (not unlike T12545). T14697 has a similar quality, being a generated multi-module test. T5837 is small enough that it similarly doesn't measure anything significant besides module loading overhead. T13253 simply does one additional round of Simplification due to Nested CPR. There are also some apparent regressions in T9198, T12234 and PmSeriesG that we (@mpickering and I) were simply unable to reproduce locally. @mpickering tried to run the CI script in a local Docker container and actually found that T9198 and PmSeriesG *improved*. In MRs that were rebased on top this one, like !4229, I did not experience such increases. Let's not get hung up on these regression tests, they were meant to test for asymptotic regressions. The build-cabal test improves by 1.2% in -O0. Metric Increase: T10421 T12234 T12545 T13035 T13056 T13701 T14697 T18923 T5837 T9198 Metric Decrease: ManyConstructors T12545 T12707 T13056 T14683 T16577 T18223 T1969 T3294 T9203 T9233 T9675 T9872a T9872b T9872c T9961 TcPlugin_RewritePerf
* Improve CSE in STG-landSimon Peyton Jones2021-04-173-0/+26
| | | | | | | | This patch fixes #19717, a long-standing bug in CSE for STG, which led to a stupid loss of CSE in some situations. It's explained in Note [Trivial case scrutinee], which I have substantially extended.
* Unit: split and rename modulesSylvain Henry2020-04-301-4/+4
| | | | | | | Introduce GHC.Unit.* hierarchy for everything concerning units, packages and modules. Update Haddock submodule
* Always do the worker/wrapper split for NOINLINEsSebastian Graf2019-03-071-9/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | Trac #10069 revealed that small NOINLINE functions didn't get split into worker and wrapper. This was due to `certainlyWillInline` saying that any unfoldings with a guidance of `UnfWhen` inline unconditionally. That isn't the case for NOINLINE functions, so we catch this case earlier now. Nofib results: -------------------------------------------------------------------------------- Program Allocs Instrs -------------------------------------------------------------------------------- fannkuch-redux -0.3% 0.0% gg +0.0% +0.1% maillist -0.2% -0.2% minimax 0.0% -0.8% -------------------------------------------------------------------------------- Min -0.3% -0.8% Max +0.0% +0.1% Geometric Mean -0.0% -0.0% Fixes #10069. ------------------------- Metric Increase: T9233 -------------------------
* testsuite: Save performance metrics in git notes.David Eichmann2018-11-071-3/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | 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
* Run StgCse after unarise, fixes #15300Ömer Sinan Ağacan2018-07-271-3/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Given two unboxed sum terms: (# 1 | #) :: (# Int | Int# #) (# 1 | #) :: (# Int | Int #) These two terms are not equal as they unarise to different unboxed tuples. However StgCse was thinking that these are equal, and replacing one of these with a binder to the other. To not deal with unboxed sums in StgCse we now do it after unarise. For StgCse to maintain post-unarise invariants we factor-out case binder in-scopeness check to `stgCaseBndrInScope` and use it in StgCse. Also did some refactoring in SimplStg. Another way to fix this would be adding a special case in StgCse to not bring unboxed sum binders in scope: diff --git a/compiler/simplStg/StgCse.hs b/compiler/simplStg/StgCse.hs index 6c740ca4cb..93a0f8f6ad 100644 --- a/compiler/simplStg/StgCse.hs +++ b/compiler/simplStg/StgCse.hs @@ -332,7 +332,11 @@ stgCseExpr env (StgLetNoEscape binds body) stgCseAlt :: CseEnv -> OutId -> InStgAlt -> OutStgAlt stgCseAlt env case_bndr (DataAlt dataCon, args, rhs) = let (env1, args') = substBndrs env args - env2 = addDataCon case_bndr dataCon (map StgVarArg args') env1 + env2 + | isUnboxedSumCon dataCon + = env1 + | otherwise + = addDataCon case_bndr dataCon (map StgVarArg args') env1 -- see note [Case 2: CSEing case binders] rhs' = stgCseExpr env2 rhs in (DataAlt dataCon, args', rhs') I think this patch seems better in that it doesn't add a special case to StgCse. Test Plan: Validate. I tried to come up with a minimal example but failed. I thought a simple program like data T = T (# Int | Int #) (# Int# | Int #) case T (# 1 | #) (# 1 | #) of ... should be enough to trigger this bug, but for some reason StgCse doesn't do anything on this program. Reviewers: simonpj, bgamari Reviewed By: simonpj Subscribers: rwbarton, thomie, carter GHC Trac Issues: #15300 Differential Revision: https://phabricator.haskell.org/D4962
* Simplify StgCases when all alts refer to the case binderJoachim Breitner2017-04-181-1/+1
| | | | | | as proposed in #13588. Differential Revision: https://phabricator.haskell.org/D3467
* Add failing test case for #13588Joachim Breitner2017-04-184-0/+230
|
* Typos in comments [ci skip]Gabor Greif2017-04-112-2/+2
|
* Add a second regression test for #13536Joachim Breitner2017-04-103-0/+38
| | | | which counts allocations instead of observing recomputation directly.
* StgCse: Do not re-use trivial case scrutineesJoachim Breitner2017-04-104-0/+20
| | | | | | | | | as they might be marked as one-shot, and suddenly we’d evaluate them multiple times. This came up in #13536 (test cases included). The solution was layed out by SPJ in ticket:13536#comment:12. Differential Revision: https://phabricator.haskell.org/D3437
* Add a CSE pass to Stg (#9291)Joachim Breitner2017-01-055-0/+81
This CSE pass only targets data constructor applications. This is probably the best we can do, as function calls and primitive operations might have side-effects. Introduces the flag -fstg-cse, enabled by default with -O for now. It might also be a good candiate for -O2. Differential Revision: https://phabricator.haskell.org/D2871
View Lorry Controller Status