[Git][ghc/ghc][wip/fix-bytecode-stubs] 16 commits: EPA: reduce [AddEpann] in AnnList

Cheng Shao (@TerrorJack) gitlab at gitlab.haskell.org
Tue Oct 29 10:49:53 UTC 2024



Cheng Shao pushed to branch wip/fix-bytecode-stubs at Glasgow Haskell Compiler / GHC


Commits:
7b1b0c6d by Alan Zimmerman at 2024-10-24T13:07:02-04:00
EPA: reduce [AddEpann] in AnnList

Remove it from the `al_rest` field, and make `AnnList` parameterized
on a type to be used in `al_rest`, for the various use cases.

- - - - -
4a00731e by Rodrigo Mesquita at 2024-10-24T13:07:38-04:00
Fix -fobject-determinism flag definition

The flag should be defined as an fflag to make sure the
-fno-object-determinism flag is also an available option.

Fixes #25397

- - - - -
55e4b9f2 by Sebastian Graf at 2024-10-25T07:01:54-04:00
CorePrep: Attach evaldUnfolding to floats to detect more values

See `Note [Pin evaluatedness on floats]`.

- - - - -
9f57c96d by Sebastian Graf at 2024-10-25T07:01:54-04:00
Make DataCon workers strict in strict fields (#20749)

This patch tweaks `exprIsConApp_maybe`, `exprIsHNF` and friends, and Demand
Analysis so that they exploit and maintain strictness of DataCon workers. See
`Note [Strict fields in Core]` for details.

Very little needed to change, and it puts field seq insertion done by Tag
Inference into a new perspective: That of *implementing* strict field semantics.
Before Tag Inference, DataCon workers are strict. Afterwards they are
effectively lazy and field seqs happen around use sites. History has shown
that there is no other way to guarantee taggedness and thus the STG Strict Field
Invariant.

Knock-on changes:

  * I reworked the whole narrative around "Tag inference".
    It's now called "EPT enforcement" and I recycyled the different overview
    Notes into `Note [EPT enforcement]`.

  * `exprIsHNF` previously used `exprOkForSpeculation` on unlifted arguments
    instead of recursing into `exprIsHNF`. That regressed the termination
    analysis in CPR analysis (which simply calls out to `exprIsHNF`), so I made
    it call `exprOkForSpeculation`, too.

  * There's a small regression in Demand Analysis, visible in the changed test
    output of T16859: Previously, a field seq on a variable would give that
    variable a "used exactly once" demand, now it's "used at least once",
    because `dmdTransformDataConSig` accounts for future uses of the field
    that actually all go through the case binder (and hence won't re-enter the
    potential thunk). The difference should hardly be observable.

  * The Simplifier's fast path for data constructors only applies to lazy
    data constructors now. I observed regressions involving Data.Binary.Put's
    `Pair` data type.

  * Unfortunately, T21392 does no longer reproduce after this patch, so I marked
    it as "not broken" in order to track whether we regress again in the future.

Fixes #20749, the satisfying conclusion of an annoying saga (cf. the ideas
in #21497 and #22475).

Compiler perf generally improves, sometimes drastically:

                                                     Baseline
                                 Test    Metric          value      New value Change
--------------------------------------------------------------------------------
             ManyConstructors(normal) ghc/alloc  3,629,760,116  3,711,852,800  +2.3%  BAD
  MultiLayerModulesTH_OneShot(normal) ghc/alloc  2,502,735,440  2,565,282,888  +2.5%  BAD
                       T12707(normal) ghc/alloc    804,399,798    791,807,320  -1.6% GOOD
                       T17516(normal) ghc/alloc    964,987,744  1,008,383,520  +4.5%
                       T18140(normal) ghc/alloc     75,381,152     49,860,560 -33.9% GOOD
                      T18698b(normal) ghc/alloc    232,614,457    184,262,736 -20.8% GOOD
                       T18923(normal) ghc/alloc     62,002,368     58,301,408  -6.0% GOOD
                       T20049(normal) ghc/alloc     75,719,168     70,494,368  -6.9% GOOD
                        T3294(normal) ghc/alloc  1,237,925,833  1,157,638,992  -6.5% GOOD
                        T9233(normal) ghc/alloc    686,490,105    635,166,688  -7.5% GOOD

                            geo. mean                                          -0.7%
                            minimum                                           -33.9%
                            maximum                                            +4.5%

I looked at T17516. It seems we do a few more simplifier iterations and end up
with a larger program. It seems that some things inline more, while other things
inline less. I don't see low-hanging fruit.

I also looked at MultiLayerModulesTH_OneShot. It appears we generate a strange
join point in the `getUnique` method of `Uniquable GHC.Unit.Types.Module` that
should better call-site inline, but does not. Perhaps with !11492.

NoFib does not seem affected much either:

+-------------------------------++--+------------+-----------+---------------+-----------+
|                               ||  |      base/ | std. err. | T20749/ (rel) | std. err. |
+===============================++==+============+===========+===============+===========+
|           spectral/last-piece ||  |    7.263e8 |      0.0% |        +0.62% |      0.0% |
+===============================++==+============+===========+===============+===========+
|                     geom mean ||  |     +0.00% |           |               |           |
+-------------------------------++--+------------+-----------+---------------+-----------+

I had a look at last-piece. Nothing changes in stg-final, but there is a bit
of ... movement around Data.Map.insert's use of GHC.Exts.lazy that is gone in
stg-final.

Co-Authored-By: Jaro Reinders <jaro.reinders at gmail.com>

Metric Decrease:
    T12707
    T18140
    T18698b
    T18923
    T19695
    T20049
    T3294
    T9233
    T21839c
Metric Increase:
    ManyConstructors
    MultiLayerModulesTH_OneShot

- - - - -
0225249a by Simon Peyton Jones at 2024-10-25T07:02:32-04:00
Some renaming

This is a pure refactor, tidying up some inconsistent naming:

   isEqPred          -->  isEqClassPred
   isEqPrimPred      -->  isEqPred
   isReprEqPrimPred  -->  isReprEqPred
   mkPrimEqPred      -->  mkNomEqPred
   mkReprPrimEqPred  -->  mkReprEqPred
   mkPrimEqPredRold  -->  mkEqPredRole

Plus I moved mkNomEqPred, mkReprEqPred, mkEqPredRolek
  from GHC.Core.Coercion to GHC.Core.Predicate
where they belong.  That means that Coercion imports Predicate
rather than vice versa -- better.

- - - - -
15a3456b by Ryan Hendrickson at 2024-10-25T07:02:32-04:00
compiler: Fix deriving with method constraints

See Note [Inferred contexts from method constraints]

Co-authored-by: Simon Peyton Jones <simon.peytonjones at gmail.com>

- - - - -
dbc77ce8 by Alan Zimmerman at 2024-10-25T18:20:13+01:00
EPA: Remove AddEpann commit 7

EPA: Remove [AddEpAnn] from HYPHEN in Parser.y

The return value is never used, as it is part of the backpack
configuration parsing.

EPA: Remove last [AddEpAnn] usages

Remove residual usage in GHC. It is still used
- In haddock TTG extension point definitions (to be removed)
- Some check-exact residual, to be removed
- Comments around DisambECP in PostProcess

EPA: Clean up [AddEpAnn] from check-exact

There is one left, to be cleaned up when we remove AddEpann itself

EPA: Remove [AddEpAnn] from haddock

The TTG extension points need a value, it is not critical what that
value is, in most cases.

EPA: Remove AddEpAnn from HsRuleAnn

EPA: Remove AddEpAnn from HsCmdArrApp

- - - - -
23ddcc01 by Simon Peyton Jones at 2024-10-26T12:44:34-04:00
Fix optimisation of InstCo

It turned out (#25387) that the fix to #15725 was not quite right:

  commit 48efbc04bd45d806c52376641e1a7ed7278d1ec7
  Date:   Mon Oct 15 10:25:02 2018 +0200

    Fix #15725 with an extra Sym

Optimising InstCo is quite subtle, and the invariants surrounding
the LiftingContext in the coercion optimiser were not stated explicitly.

This patch refactors the InstCo optimisation, and documents these
invariants.  See
  * Note [Optimising InstCo]
  * Note [The LiftingContext in optCoercion]

I also did some refactoring of course:

* Instead of a Bool swap-flag, I am not using GHC.Types.Basic.SwapFlag

* I added some invariant-checking the coercion-construction functions
  in GHC.Core.Coercion.Opt.  (Sadly these invariants don't hold during
  typechecking, becuase the types are un-zonked, so I can't put these
  checks in GHC.Core.Coercion.)

- - - - -
589fea7f by Cheng Shao at 2024-10-27T05:36:38-04:00
ghcid: use multi repl for ghcid

- - - - -
d52a0475 by Andrew Lelechenko at 2024-10-27T05:37:13-04:00
documentation: add motivating section to Control.Monad.Fix

- - - - -
301c3b54 by Cheng Shao at 2024-10-27T05:37:49-04:00
wasm: fix safari console error message related to import("node:timers")

This patch fixes the wasm backend JSFFI prelude script to avoid
calling `import("node:timers")` on non-deno hosts. Safari doesn't like
it and would print an error message to the console. Fixes
https://gitlab.haskell.org/ghc/ghc-wasm-meta/-/issues/13.

- - - - -
9f02dfb5 by Simon Peyton Jones at 2024-10-27T15:10:08-04:00
Add a missing tidy in UnivCo

We were failing to tidy the argument coercions of a UnivCo, which
led directly to #25391.

The fix is, happily, trivial.

I don't have a small repro case (it came up when building horde-ad,
which uses typechecker plugins).  It should be possible to make a
repro case, by using a plugin (which builds a UnivCo) but I decided
it was not worth the bother. The bug is egregious and easily fixed.

- - - - -
853050c3 by Andrew Lelechenko at 2024-10-27T15:10:44-04:00
Bump text submodule to 2.1.2

- - - - -
90746a59 by Andrew Lelechenko at 2024-10-27T15:10:44-04:00
hadrian: allow -Wunused-imports for text package

- - - - -
91b6f6cf by Cheng Shao at 2024-10-29T10:13:47+00:00
testsuite: add T25414 test case marked as broken

This commit adds T25414 test case to demonstrate #25414. It is marked
as broken and will be fixed by the next commit.

- - - - -
c7900ac6 by Cheng Shao at 2024-10-29T10:49:34+00:00
driver: fix foreign stub handling logic in hscParsedDecls

This patch fixes foreign stub handling logic in `hscParsedDecls`.
Previously foreign stubs were simply ignored here, so any feature that
involve foreign stubs would not work in ghci (e.g. CApiFFI). The patch
reuses `generateByteCode` logic and eliminates a large chunk of
duplicate logic that implements Core to bytecode generation pipeline
here. Fixes #25414.

- - - - -


30 changed files:

- .ghcid
- compiler/GHC/Builtin/Types.hs
- compiler/GHC/Builtin/Types/Prim.hs
- compiler/GHC/Core.hs
- compiler/GHC/Core/Coercion.hs
- compiler/GHC/Core/Coercion.hs-boot
- compiler/GHC/Core/Coercion/Opt.hs
- compiler/GHC/Core/DataCon.hs
- compiler/GHC/Core/Opt/Arity.hs
- compiler/GHC/Core/Opt/CprAnal.hs
- compiler/GHC/Core/Opt/DmdAnal.hs
- compiler/GHC/Core/Opt/Simplify/Env.hs
- compiler/GHC/Core/Opt/Simplify/Iteration.hs
- compiler/GHC/Core/Opt/WorkWrap/Utils.hs
- compiler/GHC/Core/Predicate.hs
- compiler/GHC/Core/SimpleOpt.hs
- compiler/GHC/Core/TyCo/Subst.hs
- compiler/GHC/Core/TyCo/Tidy.hs
- compiler/GHC/Core/Type.hs
- compiler/GHC/Core/Utils.hs
- compiler/GHC/CoreToStg/Prep.hs
- compiler/GHC/Driver/Main.hs
- compiler/GHC/Driver/Session.hs
- compiler/GHC/Hs/Binds.hs
- compiler/GHC/Hs/Decls.hs
- compiler/GHC/Hs/Dump.hs
- compiler/GHC/Hs/Expr.hs
- compiler/GHC/Hs/ImpExp.hs
- compiler/GHC/Hs/Pat.hs
- compiler/GHC/Hs/Type.hs


The diff was not included because it is too large.


View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/677e3aa56e905524071fc9717a88ad2cd1bc2951...c7900ac6e1bf5288f03ba388f124ceafd9d79b6f

-- 
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/677e3aa56e905524071fc9717a88ad2cd1bc2951...c7900ac6e1bf5288f03ba388f124ceafd9d79b6f
You're receiving this email because of your account on gitlab.haskell.org.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-commits/attachments/20241029/d8cdfea8/attachment-0001.html>


More information about the ghc-commits mailing list