[Git][ghc/ghc][wip/T18078] 23 commits: GHC.Core.Unfold: Refactor traceInline
Simon Peyton Jones
gitlab at gitlab.haskell.org
Fri May 29 22:15:42 UTC 2020
Simon Peyton Jones pushed to branch wip/T18078 at Glasgow Haskell Compiler / GHC
Commits:
28deee28 by Ben Gamari at 2020-05-28T16:23:21-04:00
GHC.Core.Unfold: Refactor traceInline
This reduces duplication as well as fixes a bug wherein -dinlining-check
would override -ddump-inlinings. Moreover, the new variant
- - - - -
1f393e1e by Ben Gamari at 2020-05-28T16:23:21-04:00
Avoid unnecessary allocations due to tracing utilities
While ticky-profiling the typechecker I noticed that hundreds of
millions of SDocs are being allocated just in case -ddump-*-trace is
enabled. This is awful.
We avoid this by ensuring that the dump flag check is inlined into the
call site, ensuring that the tracing document needn't be allocated
unless it's actually needed.
See Note [INLINE conditional tracing utilities] for details.
Fixes #18168.
Metric Decrease:
T9961
haddock.Cabal
haddock.base
haddock.compiler
- - - - -
5f621a78 by Vladislav Zavialov at 2020-05-28T16:23:58-04:00
Add Semigroup/Monoid for Q (#18123)
- - - - -
dc5f004c by Xavier Denis at 2020-05-28T16:24:37-04:00
Fix #18071
Run the core linter on candidate instances to ensure they are
well-kinded.
Better handle quantified constraints by using a CtWanted to avoid
having unsolved constraints thrown away at the end by the solver.
- - - - -
10e6982c by Sebastian Graf at 2020-05-28T16:25:14-04:00
FloatOut: Only eta-expand dead-end RHS if arity will increase (#18231)
Otherwise we risk turning trivial RHS into non-trivial RHS, introducing
unnecessary bindings in the next Simplifier run, resulting in more
churn.
Fixes #18231.
- - - - -
08dab5f7 by Sebastian Graf at 2020-05-28T16:25:14-04:00
DmdAnal: Recognise precise exceptions from case alternatives (#18086)
Consider
```hs
m :: IO ()
m = do
putStrLn "foo"
error "bar"
```
`m` (from #18086) always throws a (precise or imprecise) exception or
diverges. Yet demand analysis infers `<L,A>` as demand signature instead
of `<L,A>x` for it.
That's because the demand analyser sees `putStrLn` occuring in a case
scrutinee and decides that it has to `deferAfterPreciseException`,
because `putStrLn` throws a precise exception on some control flow
paths. This will mask the `botDiv` `Divergence`of the single case alt
containing `error` to `topDiv`. Since `putStrLn` has `topDiv` itself,
the final `Divergence` is `topDiv`.
This is easily fixed: `deferAfterPreciseException` works by `lub`ing
with the demand type of a virtual case branch denoting the precise
exceptional control flow. We used `nopDmdType` before, but we can be
more precise and use `exnDmdType`, which is `nopDmdType` with `exnDiv`.
Now the `Divergence` from the case alt will degrade `botDiv` to `exnDiv`
instead of `topDiv`, which combines with the result from the scrutinee
to `exnDiv`, and all is well.
Fixes #18086.
- - - - -
aef95f11 by Ben Gamari at 2020-05-28T16:25:53-04:00
Ticky-ticky: Record DataCon name in ticker name
This makes it significantly easier to spot the nature of
allocations regressions and comes at a reasonably low cost.
- - - - -
8f021b8c by Ben Gamari at 2020-05-28T16:26:34-04:00
hadrian: Don't track GHC's verbosity argument
Teach hadrian to ignore GHC's -v argument in its recompilation check,
thus fixing #18131.
- - - - -
13d9380b by Ben Gamari at 2020-05-28T16:27:20-04:00
Rip out CmmStackInfo(updfr_space)
As noted in #18232, this field is currently completely unused and
moreover doesn't have a clear meaning.
- - - - -
f10d11fa by Andreas Klebinger at 2020-05-29T01:38:42-04:00
Fix "build/elem" RULE.
An redundant constraint prevented the rule from matching.
Fixing this allows a call to elem on a known list to be translated
into a series of equality checks, and eventually a simple case
expression.
Surprisingly this seems to regress elem for strings. To avoid
this we now also allow foldrCString to inline and add an UTF8
variant. This results in elem being compiled to a tight
non-allocating loop over the primitive string literal which
performs a linear search.
In the process this commit adds UTF8 variants for some of the
functions in GHC.CString. This is required to make this work for
both ASCII and UTF8 strings.
There are also small tweaks to the CString related rules.
We now allow ourselfes the luxury to compare the folding function
via eqExpr, which helps to ensure the rule fires before we inline
foldrCString*. Together with a few changes to allow matching on both
the UTF8 and ASCII variants of the CString functions.
- - - - -
bbeb2389 by Ben Gamari at 2020-05-29T01:39:19-04:00
CoreToStg: Add Outputable ArgInfo instance
- - - - -
0e3361ca by Simon Peyton Jones at 2020-05-29T01:39:19-04:00
Make Lint check return type of a join point
Consider
join x = rhs in body
It's important that the type of 'rhs' is the same as the type of
'body', but Lint wasn't checking that invariant.
Now it does! This was exposed by investigation into !3113.
- - - - -
c49f7df0 by Simon Peyton Jones at 2020-05-29T01:39:19-04:00
Do not float join points in exprIsConApp_maybe
We hvae been making exprIsConApp_maybe cleverer in recent times:
commit b78cc64e923716ac0512c299f42d4d0012306c05
Date: Thu Nov 15 17:14:31 2018 +0100
Make constructor wrappers inline only during the final phase
commit 7833cf407d1f608bebb1d38bb99d3035d8d735e6
Date: Thu Jan 24 17:58:50 2019 +0100
Look through newtype wrappers (Trac #16254)
commit c25b135ff5b9c69a90df0ccf51b04952c2dc6ee1
Date: Thu Feb 21 12:03:22 2019 +0000
Fix exprIsConApp_maybe
But alas there was still a bug, now immortalised in
Note [Don't float join points]
in SimpleOpt.
It's quite hard to trigger because it requires a dead
join point, but it came up when compiling Cabal
Cabal.Distribution.Fields.Lexer.hs, when working on
!3113.
Happily, the fix is extremly easy. Finding the
bug was not so easy.
- - - - -
46720997 by Ben Gamari at 2020-05-29T01:39:19-04:00
Allow simplification through runRW#
Because runRW# inlines so late, we were previously able to do very
little simplification across it. For instance, given even a simple
program like
case runRW# (\s -> let n = I# 42# in n) of
I# n# -> f n#
we previously had no way to avoid the allocation of the I#.
This patch allows the simplifier to push strict contexts into the
continuation of a runRW# application, as explained in
in Note [Simplification of runRW#] in GHC.CoreToStg.Prep.
Fixes #15127.
Metric Increase:
T9961
Metric Decrease:
ManyConstructors
Co-Authored-By: Simon Peyton-Jone <simonpj at microsoft.com>
- - - - -
277c2f26 by Ben Gamari at 2020-05-29T01:39:55-04:00
Eta expand un-saturated primops
Now since we no longer try to predict CAFfyness we have no need for the
solution to #16846. Eta expanding unsaturated primop applications is
conceptually simpler, especially in the presence of levity polymorphism.
This essentially reverts cac8dc9f51e31e4c0a6cd9bc302f7e1bc7c03beb,
as suggested in #18079.
Closes #18079.
- - - - -
f44d7ae0 by Simon Jakobi at 2020-05-29T01:40:34-04:00
base: Scrap deprecation plan for Data.Monoid.{First,Last}
See the discussion on the libraries mailing list for context:
https://mail.haskell.org/pipermail/libraries/2020-April/030357.html
- - - - -
8b494895 by Jeremy Schlatter at 2020-05-29T01:41:12-04:00
Fix typo in documentation
- - - - -
998450f4 by Gleb Popov at 2020-05-29T01:41:53-04:00
Always define USE_PTHREAD_FOR_ITIMER for FreeBSD.
- - - - -
f9a513e0 by Alp Mestanogullari at 2020-05-29T01:42:36-04:00
hadrian: introduce 'install' target
Its logic is very simple. It `need`s the `binary-dist-dir` target
and runs suitable `configure` and `make install` commands for the
user. A new `--prefix` command line argument is introduced to
specify where GHC should be installed.
- - - - -
67738db1 by Travis Whitaker at 2020-05-29T13:34:48-04:00
Build a threaded stage 1 if the bootstrapping GHC supports it.
- - - - -
aac19e6c by Peter Trommler at 2020-05-29T13:35:24-04:00
PPC NCG: No per-symbol .section ".toc" directives
All position independent symbols are collected during code generation
and emitted in one go. Prepending each symbol with a .section ".toc"
directive is redundant. This patch drops the per-symbol directives
leading to smaller assembler files.
Fixes #18250
- - - - -
c562c313 by Simon Peyton Jones at 2020-05-29T23:02:20+01:00
Optimisation in Unique.Supply
This patch switches on -fno-state-hack in GHC.Types.Unique.Supply.
It turned out that my fixes for #18078 (coercion floating) changed the
optimisation pathway for mkSplitUniqSupply in such a way that we had
an extra allocation inside the inner loop. Adding -fno-state-hack
fixed that -- and indeed the loop in mkSplitUniqSupply is a classic
example of the way in which -fno-state-hack can be bad; see #18238.
Moreover, the new code is better than the old. They allocate
the same, but the old code ends up with a partial application.
The net effect is that the test
perf/should_run/UniqLoop
runs 20% faster! From 2.5s down to 2.0s. The allocation numbers
are the same -- but elapsed time falls. Good!
The bad thing about this is that it's terribly delicate. But
at least it's a good example of such delicacy in action.
There is a long Note [Optimising the unique supply] which now
explains all this.
- - - - -
12f458eb by Simon Peyton Jones at 2020-05-29T23:05:06+01:00
Implement cast worker/wrapper properly
The cast worker/wrapper transformation transforms
x = e |> co
into
y = e
x = y |> co
This is done by the simplifier, but we were being
careless about transferring IdInfo from x to y,
and about what to do if x is a NOINLNE function.
This resulted in a series of bugs:
#17673, #18093, #18078.
This patch fixes all that:
* Main change is in GHC.Core.Opt.Simplify, and
the new prepareBinding function, which does this
cast worker/wrapper transform.
See Note [Cast worker/wrappers].
* There is quite a bit of refactoring around
prepareRhs, makeTrivial etc. It's nicer now.
* Some wrappers from strictness and cast w/w, notably those for
a function with a NOINLINE, should inline very late. There
wasn't really a mechanism for that, which was an existing bug
really; so I invented a new finalPhase = Phase (-1). It's used
for all simplifier runs after the user-visible phase 2,1,0 have
run. (No new runs of the simplifier are introduced thereby.)
See new Note [Compiler phases] in GHC.Types.Basic;
the main changes are in GHC.Core.Opt.Driver
* Doing this made me trip over two places where the AnonArgFlag on a
FunTy was being lost so we could end up with (Num a -> ty)
rather than (Num a => ty)
- In coercionLKind/coercionRKind
- In contHoleType in the Simplifier
I fixed the former by defining mkFunctionType and using it in
coercionLKind/RKind.
I could have done the same for the latter, but the information
is almost to hand. So I fixed the latter by
- adding sc_hole_ty to ApplyToVal (like ApplyToTy),
- adding as_hole_ty to ValArg (like TyArg)
- adding sc_fun_ty to StrictArg
Turned out I could then remove ai_type from ArgInfo. This is
just moving the deck chairs around, but it worked out nicely.
See the new Note [AnonArgFlag] in GHC.Types.Var
* When looking at the 'arity decrease' thing (#18093) I discovered
that stable unfoldings had a much lower arity than the actual
optimised function. That's what led to the arity-decrease
message. Simple solution: eta-expand.
It's described in Note [Eta-expand stable unfoldings]
in GHC.Core.Opt.Simplify
* I also discovered that unsafeCoerce wasn't being inlined if
the context was boring. So (\x. f (unsafeCoerce x)) would
create a thunk -- yikes! I fixed that by making inlineBoringOK
a bit cleverer: see Note [Inline unsafeCoerce] in GHC.Core.Unfold.
I also found that unsafeCoerceName was unused, so I removed it.
I made a test case for #18078, and a very similar one for #17673.
The net effect of all this on nofib is very modest, but positive:
--------------------------------------------------------------------------------
Program Size Allocs Runtime Elapsed TotalMem
--------------------------------------------------------------------------------
anna -0.4% -0.1% -3.1% -3.1% 0.0%
fannkuch-redux -0.4% -0.3% -0.1% -0.1% 0.0%
maillist -0.4% -0.1% -7.8% -1.0% -14.3%
primetest -0.4% -15.6% -7.1% -6.6% 0.0%
--------------------------------------------------------------------------------
Min -0.9% -15.6% -13.3% -14.2% -14.3%
Max -0.3% 0.0% +12.1% +12.4% 0.0%
Geometric Mean -0.4% -0.2% -2.3% -2.2% -0.1%
Metric Decrease:
T15164
T13701
Metric Increase:
T12150
T12234
T12425
- - - - -
30 changed files:
- compiler/GHC/Builtin/Names.hs
- compiler/GHC/Builtin/PrimOps.hs
- compiler/GHC/ByteCode/Linker.hs
- compiler/GHC/Cmm.hs
- compiler/GHC/Cmm/LayoutStack.hs
- compiler/GHC/Cmm/Ppr.hs
- compiler/GHC/Cmm/ProcPoint.hs
- compiler/GHC/CmmToAsm/PIC.hs
- compiler/GHC/Core/Coercion.hs
- compiler/GHC/Core/Lint.hs
- compiler/GHC/Core/Opt/CSE.hs
- compiler/GHC/Core/Opt/ConstantFold.hs
- compiler/GHC/Core/Opt/CprAnal.hs
- compiler/GHC/Core/Opt/Driver.hs
- compiler/GHC/Core/Opt/FloatOut.hs
- compiler/GHC/Core/Opt/OccurAnal.hs
- compiler/GHC/Core/Opt/SetLevels.hs
- compiler/GHC/Core/Opt/Simplify.hs
- compiler/GHC/Core/Opt/Simplify/Monad.hs
- compiler/GHC/Core/Opt/Simplify/Utils.hs
- compiler/GHC/Core/Opt/SpecConstr.hs
- compiler/GHC/Core/Opt/WorkWrap.hs
- compiler/GHC/Core/SimpleOpt.hs
- compiler/GHC/Core/Type.hs
- compiler/GHC/Core/Unfold.hs
- compiler/GHC/Core/Utils.hs
- compiler/GHC/CoreToStg.hs
- compiler/GHC/CoreToStg/Prep.hs
- compiler/GHC/HsToCore/PmCheck/Oracle.hs
- compiler/GHC/HsToCore/Utils.hs
The diff was not included because it is too large.
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/47025bbf175086d6eec1599de5ffb9c078498145...12f458eb7e63b6ca8a2d5a3c94ab355582763ecc
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/47025bbf175086d6eec1599de5ffb9c078498145...12f458eb7e63b6ca8a2d5a3c94ab355582763ecc
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/20200529/129406a9/attachment-0001.html>
More information about the ghc-commits
mailing list