[Git][ghc/ghc][wip/T17923] 8 commits: Don't override proc CafInfos in ticky builds
Simon Peyton Jones
gitlab at gitlab.haskell.org
Wed Apr 8 08:18:02 UTC 2020
Simon Peyton Jones pushed to branch wip/T17923 at Glasgow Haskell Compiler / GHC
Commits:
dcfe29c8 by Ömer Sinan Ağacan at 2020-04-06T13:16:08-04:00
Don't override proc CafInfos in ticky builds
Fixes #17947
When we have a ticky label for a proc, IdLabels for the ticky counter
and proc entry share the same Name. This caused overriding proc CafInfos
with the ticky CafInfos (i.e. NoCafRefs) during SRT analysis.
We now ignore the ticky labels when building SRTMaps. This makes sense
because:
- When building the current module they don't need to be in SRTMaps as
they're initialized as non-CAFFY (see mkRednCountsLabel), so they
don't take part in the dependency analysis and they're never added to
SRTs.
(Reminder: a "dependency" in the SRT analysis is a CAFFY dependency,
non-CAFFY uses are not considered as dependencies for the algorithm)
- They don't appear in the interfaces as they're not exported, so it
doesn't matter for cross-module concerns whether they're in the SRTMap
or not.
See also the new Note [Ticky labels in SRT analysis].
- - - - -
cec2c71f by Simon Peyton Jones at 2020-04-06T13:16:44-04:00
Fix an tricky specialiser loop
Issue #17151 was a very tricky example of a bug in which the
specialiser accidentally constructs a recurive dictionary,
so that everything turns into bottom.
I have fixed variants of this bug at least twice before:
see Note [Avoiding loops]. It was a bit of a struggle
to isolate the problem, greatly aided by the work that
Alexey Kuleshevich did in distilling a test case.
Once I'd understood the problem, it was not difficult to fix,
though it did lead me a bit of refactoring in specImports.
- - - - -
e850d14f by Simon Peyton Jones at 2020-04-06T13:16:44-04:00
Refactoring only
This refactors DictBinds into a data type rather than a pair.
No change in behaviour, just better code
- - - - -
f38e8d61 by Daniel Gröber at 2020-04-07T02:00:05-04:00
rts: ProfHeap: Fix memory leak when not compiled with profiling
If we're doing heap profiling on an unprofiled executable we keep
allocating new space in initEra via nextEra on each profiler run but we
don't have a corresponding freeEra call.
We do free the last era in endHeapProfiling but previous eras will have
been overwritten by initEra and will never get free()ed.
Metric Decrease:
space_leak_001
- - - - -
bcd66859 by Sebastian Graf at 2020-04-07T02:00:41-04:00
Re-export GHC.Magic.noinline from base
- - - - -
3d2991f8 by Ben Gamari at 2020-04-07T18:36:09-04:00
simplifier: Kill off ufKeenessFactor
We used to have another factor, ufKeenessFactor, which would scale the
discounts before they were subtracted from the size. This was justified
with the following comment:
-- We multiple the raw discounts (args_discount and result_discount)
-- ty opt_UnfoldingKeenessFactor because the former have to do with
-- *size* whereas the discounts imply that there's some extra
-- *efficiency* to be gained (e.g. beta reductions, case reductions)
-- by inlining.
However, this is highly suspect since it means that we subtract a
*scaled* size from an absolute size, resulting in crazy (e.g. negative)
scores in some cases (#15304). We consequently killed off
ufKeenessFactor and bumped up the ufUseThreshold to compensate.
Adjustment of unfolding use threshold
=====================================
Since this removes a discount from our inlining heuristic, I revisited our
default choice of -funfolding-use-threshold to minimize the change in
overall inlining behavior. Specifically, I measured runtime allocations
and executable size of nofib and the testsuite performance tests built
using compilers (and core libraries) built with several values of
-funfolding-use-threshold.
This comes as a result of a quantitative comparison of testsuite
performance and code size as a function of ufUseThreshold, comparing
GHC trees using values of 50, 60, 70, 80, 90, and 100. The test set
consisted of nofib and the testsuite performance tests.
A full summary of these measurements are found in the description of
!2608
Comparing executable sizes (relative to the base commit) across all
nofib tests, we see that sizes are similar to the baseline:
gmean min max median
thresh
50 -6.36% -7.04% -4.82% -6.46%
60 -5.04% -5.97% -3.83% -5.11%
70 -2.90% -3.84% -2.31% -2.92%
80 -0.75% -2.16% -0.42% -0.73%
90 +0.24% -0.41% +0.55% +0.26%
100 +1.36% +0.80% +1.64% +1.37%
baseline +0.00% +0.00% +0.00% +0.00%
Likewise, looking at runtime allocations we see that 80 gives slightly
better optimisation than the baseline:
gmean min max median
thresh
50 +0.16% -0.16% +4.43% +0.00%
60 +0.09% -0.00% +3.10% +0.00%
70 +0.04% -0.09% +2.29% +0.00%
80 +0.02% -1.17% +2.29% +0.00%
90 -0.02% -2.59% +1.86% +0.00%
100 +0.00% -2.59% +7.51% -0.00%
baseline +0.00% +0.00% +0.00% +0.00%
Finally, I had to add a NOINLINE in T4306 to ensure that `upd` is
worker-wrappered as the test expects. This makes me wonder whether the
inlining heuristic is now too liberal as `upd` is quite a large
function. The same measure was taken in T12600.
Wall clock time compiling Cabal with -O0
thresh 50 60 70 80 90 100 baseline
build-Cabal 93.88 89.58 92.59 90.09 100.26 94.81 89.13
Also, this change happens to avoid the spurious test output in
`plugin-recomp-change` and `plugin-recomp-change-prof` (see #17308).
Metric Decrease:
hie002
T12234
T13035
T13719
T14683
T4801
T5631
T5642
T9020
T9872d
T9961
Metric Increase:
T12150
T12425
T13701
T14697
T15426
T1969
T3064
T5837
T6048
T9203
T9872a
T9872b
T9872c
T9872d
haddock.Cabal
haddock.base
haddock.compiler
- - - - -
255418da by Sylvain Henry at 2020-04-07T18:36:49-04:00
Modules: type-checker (#13009)
Update Haddock submodule
- - - - -
582ee21f by Simon Peyton Jones at 2020-04-08T09:17:24+01:00
Significant refactor of Lint
This refactoring of Lint was triggered by #17923, which is
fixed by this patch.
The main change is this. Instead of
lintType :: Type -> LintM LintedKind
we now have
lintType :: Type -> LintM LintedType
Previously, all of typeKind was effectively duplicate in lintType.
Moreover, since we have an ambient substitution, we still had to
apply the substition here and there, sometimes more than once. It
was all very tricky, in the end, and made my head hurt.
Now, lintType returns a fully linted type, with all substitutions
performed on it. This is much simpler.
The same thing is needed for Coercions. Instead of
lintCoercion :: OutCoercion
-> LintM (LintedKind, LintedKind,
LintedType, LintedType, Role)
we now have
lintCoercion :: Coercion -> LintM LintedCoercion
Much simpler! The code is shorter and less bug-prone.
There are a lot of knock on effects. But life is now better.
Metric Decrease:
T1969
- - - - -
30 changed files:
- CODEOWNERS
- compiler/GHC.hs
- compiler/GHC/Cmm/CLabel.hs
- compiler/GHC/Cmm/Expr.hs
- compiler/GHC/Cmm/Info/Build.hs
- compiler/GHC/Cmm/Node.hs
- compiler/GHC/Core.hs
- compiler/GHC/Core/Arity.hs
- compiler/GHC/Core/Class.hs
- compiler/GHC/Core/Coercion.hs
- compiler/GHC/Core/Coercion/Axiom.hs
- compiler/GHC/Core/Coercion/Opt.hs
- compiler/GHC/Core/FamInstEnv.hs
- compiler/GHC/Core/InstEnv.hs
- compiler/GHC/Core/Lint.hs
- compiler/GHC/Core/Op/CSE.hs
- compiler/GHC/Core/Op/OccurAnal.hs
- compiler/GHC/Core/Op/Simplify.hs
- compiler/GHC/Core/Op/Specialise.hs
- compiler/GHC/Core/PatSyn.hs
- compiler/GHC/Core/Predicate.hs
- compiler/GHC/Core/Rules.hs
- compiler/GHC/Core/TyCo/FVs.hs
- compiler/GHC/Core/TyCo/Ppr.hs
- compiler/GHC/Core/TyCo/Rep.hs
- compiler/GHC/Core/TyCon.hs
- compiler/GHC/Core/Type.hs
- compiler/GHC/Core/Unfold.hs
- compiler/GHC/Core/Unify.hs
- compiler/GHC/CoreToStg/Prep.hs
The diff was not included because it is too large.
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/92904edf42b8bf154e3e6ab0680bb17d31d0d71b...582ee21f82eadac2d3e92d07f4f0ac4e241e31d3
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/92904edf42b8bf154e3e6ab0680bb17d31d0d71b...582ee21f82eadac2d3e92d07f4f0ac4e241e31d3
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/20200408/dda7ce33/attachment.html>
More information about the ghc-commits
mailing list