[Git][ghc/ghc][wip/marge_bot_batch_merge_job] 6 commits: Better eta-expansion (again) and don't specilise DFuns

Marge Bot gitlab at gitlab.haskell.org
Tue Sep 22 21:12:03 UTC 2020 


 Marge Bot pushed to branch wip/marge_bot_batch_merge_job at Glasgow Haskell Compiler / GHC


Commits:
6de40f83 by Simon Peyton Jones at 2020-09-22T05:37:24-04:00
Better eta-expansion (again) and don't specilise DFuns

This patch fixes #18223, which made GHC generate an exponential
amount of code.  There are three quite separate changes in here

1.  Re-engineer eta-expansion (again).  The eta-expander was
    generating lots of intermediate stuff, which could be optimised
    away, but which choked the simplifier meanwhile.  Relatively
    easy to kill it off at source.

    See Note [The EtaInfo mechanism] in GHC.Core.Opt.Arity.
    The main new thing is the use of pushCoArg in getArg_maybe.

2.  Stop Specialise specalising DFuns.  This is the cause of a huge
    (and utterly unnecessary) blowup in program size in #18223.
    See Note [Do not specialise DFuns] in GHC.Core.Opt.Specialise.

    I also refactored the Specialise monad a bit... it was silly,
    because it passed on unchanging values as if they were mutable
    state.

3.  Do an extra Simplifer run, after SpecConstra and before
    late-Specialise.  I found (investigating perf/compiler/T16473)
    that failing to do this was crippling *both* SpecConstr *and*
    Specialise.  See Note [Simplify after SpecConstr] in
    GHC.Core.Opt.Pipeline.

    This change does mean an extra run of the Simplifier, but only
    with -O2, and I think that's acceptable.

    T16473 allocates *three* times less with this change.  (I changed
    it to check runtime rather than compile time.)

Some smaller consequences

* I moved pushCoercion, pushCoArg and friends from SimpleOpt
  to Arity, because it was needed by the new etaInfoApp.

  And pushCoValArg now returns a MCoercion rather than Coercion for
  the argument Coercion.

* A minor, incidental improvement to Core pretty-printing

This does fix #18223, (which was otherwise uncompilable. Hooray.  But
there is still a big intermediate because there are some very deeply
nested types in that program.

Modest reductions in compile-time allocation on a couple of benchmarks
    T12425     -2.0%
    T13253    -10.3%

Metric increase with -O2, due to extra simplifier run
    T9233     +5.8%
    T12227    +1.8%
    T15630    +5.0%

There is a spurious apparent increase on heap residency on T9630,
on some architectures at least.  I tried it with -G1 and the residency
is essentially unchanged.

Metric Increase
    T9233
    T12227
    T9630

Metric Decrease
    T12425
    T13253

- - - - -
416bd50e by Simon Peyton Jones at 2020-09-22T05:37:59-04:00
Fix the occurrence analyser

Ticket #18603 demonstrated that the occurrence analyser's
handling of

  local RULES for imported Ids

(which I now call IMP-RULES) was inadequate.  It led the simplifier
into an infnite loop by failing to label a binder as a loop breaker.

The main change in this commit is to treat IMP-RULES in a simple and
uniform way: as extra rules for the local binder.  See
  Note [IMP-RULES: local rules for imported functions]

This led to quite a bit of refactoring.  The result is still tricky,
but it's much better than before, and better documented I think.

Oh, and it fixes the bug.

- - - - -
6fe8a0c7 by Sebastian Graf at 2020-09-22T05:38:35-04:00
PmCheck - Comments only: Replace /~ by ≁

- - - - -
e9501547 by Sebastian Graf at 2020-09-22T05:38:35-04:00
PmCheck: Rewrite inhabitation test

We used to produce inhabitants of a pattern-match refinement type Nabla
in the checker in at least two different and mostly redundant ways:

  1. There was `provideEvidence` (now called
     `generateInhabitingPatterns`) which is used by
     `GHC.HsToCore.PmCheck` to produce non-exhaustive patterns, which
     produces inhabitants of a Nabla as a sub-refinement type where all
     match variables are instantiated.
  2. There also was `ensure{,All}Inhabited` (now called
     `inhabitationTest`) which worked slightly different, but was
     whenever new type constraints or negative term constraints were
     added. See below why `provideEvidence` and `ensureAllInhabited`
     can't be the same function, the main reason being performance.
  3. And last but not least there was the `nonVoid` test, which tested
     that a given type was inhabited. We did use this for strict fields
     and -XEmptyCase in the past.

The overlap of (3) with (2) was always a major pet peeve of mine. The
latter was quite efficient and proven to work for recursive data types,
etc, but could not handle negative constraints well (e.g. we often want
to know if a *refined* type is empty, such as `{ x:[a] | x /= [] }`).

Lower Your Guards suggested that we could get by with just one, by
replacing both functions with `inhabitationTest` in this patch.
That was only possible by implementing the structure of φ constraints
as in the paper, namely the semantics of φ constructor constraints.

This has a number of benefits:

  a. Proper handling of unlifted types and strict fields, fixing #18249,
     without any code duplication between
     `GHC.HsToCore.PmCheck.Oracle.instCon` (was `mkOneConFull`) and
     `GHC.HsToCore.PmCheck.checkGrd`.
  b. `instCon` can perform the `nonVoid` test (3) simply by emitting
     unliftedness constraints for strict fields.
  c. `nonVoid` (3) is thus simply expressed by a call to
     `inhabitationTest`.
  d. Similarly, `ensureAllInhabited` (2), which we called after adding
     type info, now can similarly be expressed as the fuel-based
     `inhabitationTest`.

See the new `Note [Why inhabitationTest doesn't call generateInhabitingPatterns]`
why we still have tests (1) and (2).

Fixes #18249 and brings nice metric decreases for `T17836` (-76%) and
`T17836b` (-46%), as well as `T18478` (-8%) at the cost of a few very
minor regressions (< +2%), potentially due to the fact that
`generateInhabitingPatterns` does more work to suggest the minimal
COMPLETE set.

Metric Decrease:
    T17836
    T17836b

- - - - -
63ef941a by Hécate at 2020-09-22T17:11:54-04:00
Remove the list of loaded modules from the ghci prompt

- - - - -
0a217d1d by Ben Gamari at 2020-09-22T17:11:54-04:00
Bump submodules

* Bump bytestring to 0.10.12.0
* Bump Cabal to 3.4.0.0-rc3
* Bump Win32 to 2.10.0.0

- - - - -


19 changed files:

- compiler/GHC/Core/Coercion.hs
- compiler/GHC/Core/FVs.hs
- compiler/GHC/Core/Opt/Arity.hs
- compiler/GHC/Core/Opt/OccurAnal.hs
- compiler/GHC/Core/Opt/Pipeline.hs
- compiler/GHC/Core/Opt/Simplify.hs
- compiler/GHC/Core/Opt/Simplify/Env.hs
- compiler/GHC/Core/Opt/Specialise.hs
- compiler/GHC/Core/Ppr.hs
- compiler/GHC/Core/SimpleOpt.hs
- compiler/GHC/Core/Subst.hs
- compiler/GHC/Hs/Expr.hs
- compiler/GHC/HsToCore/PmCheck.hs
- compiler/GHC/HsToCore/PmCheck/Oracle.hs
- compiler/GHC/HsToCore/PmCheck/Ppr.hs
- compiler/GHC/HsToCore/PmCheck/Types.hs
- − compiler/GHC/HsToCore/PmCheck/Types.hs-boot
- compiler/GHC/HsToCore/Types.hs
- compiler/GHC/Tc/Gen/Expr.hs


The diff was not included because it is too large.


View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/e9abf8daca866c326195586a19be3477a5d0a0d4...0a217d1de3de01168b33e2dde05dcc6e5ffef6c2

-- 
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/e9abf8daca866c326195586a19be3477a5d0a0d4...0a217d1de3de01168b33e2dde05dcc6e5ffef6c2
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/20200922/04e740a2/attachment-0001.html>

More information about the ghc-commits mailing list