[Git][ghc/ghc][wip/t23103] 35 commits: configure: Derive library version from ghc-prim.cabal.in

Finley McIlwaine (@FinleyMcIlwaine) gitlab at gitlab.haskell.org
Mon Aug 14 14:35:19 UTC 2023 


Finley McIlwaine pushed to branch wip/t23103 at Glasgow Haskell Compiler / GHC


Commits:
01961be3 by Ben Gamari at 2023-08-08T02:47:14-04:00
configure: Derive library version from ghc-prim.cabal.in

Since ghc-prim.cabal is now generated by Hadrian, we cannot depend upon
it.

Closes #23726.

- - - - -
3b373838 by Ryan Scott at 2023-08-08T02:47:49-04:00
tcExpr: Push expected types for untyped TH splices inwards

In !10911, I deleted a `tcExpr` case for `HsUntypedSplice` in favor of a much
simpler case that simply delegates to `tcApp`. Although this passed the test
suite at the time, this was actually an error, as the previous `tcExpr` case
was critically pushing the expected type inwards. This actually matters for
programs like the one in #23796, which GHC would not accept with type inference
alone—we need full-blown type _checking_ to accept these.

I have added back the previous `tcExpr` case for `HsUntypedSplice` and now
explain why we have two different `HsUntypedSplice` cases (one in `tcExpr` and
another in `splitHsApps`) in `Note [Looking through Template Haskell splices in
splitHsApps]` in `GHC.Tc.Gen.Head`.

Fixes #23796.

- - - - -
0ef1d8ae by sheaf at 2023-08-08T21:26:51-04:00
Compute all emitted diagnostic codes

This commit introduces in GHC.Types.Error.Codes the function

  constructorCodes :: forall diag. (...) => Map DiagnosticCode String

which computes a collection of all the diagnostic codes that correspond
to a particular type. In particular, we can compute the collection of
all diagnostic codes emitted by GHC using the invocation

  constructorCodes @GhcMessage

We then make use of this functionality in the new "codes" test which
checks consistency and coverage of GHC diagnostic codes.
It performs three checks:

  - check 1: all non-outdated GhcDiagnosticCode equations
    are statically used.
  - check 2: all outdated GhcDiagnosticCode equations
    are statically unused.
  - check 3: all statically used diagnostic codes are covered by
    the testsuite (modulo accepted exceptions).

- - - - -
4bc7b1e5 by Fraser Tweedale at 2023-08-08T21:27:32-04:00
numberToRangedRational: fix edge cases for exp ≈ (maxBound :: Int)

Currently a negative exponent less than `minBound :: Int` results in
Infinity, which is very surprising and obviously wrong.

```
λ> read "1e-9223372036854775808" :: Double
0.0
λ> read "1e-9223372036854775809" :: Double
Infinity
```

There is a further edge case where the exponent can overflow when
increased by the number of tens places in the integer part, or
underflow when decreased by the number of leading zeros in the
fractional part if the integer part is zero:

```
λ> read "10e9223372036854775807" :: Double
0.0
λ> read "0.01e-9223372036854775808" :: Double
Infinity
```

To resolve both of these issues, perform all arithmetic and
comparisons involving the exponent in type `Integer`.  This approach
also eliminates the need to explicitly check the exponent against
`maxBound :: Int` and `minBound :: Int`, because the allowed range
of the exponent (i.e. the result of `floatRange` for the target
floating point type) is certainly within those bounds.

This change implements CLC proposal 192:
https://github.com/haskell/core-libraries-committee/issues/192

- - - - -
6eab07b2 by Alan Zimmerman at 2023-08-08T21:28:10-04:00
EPA: Remove Location from WarningTxt source

This is not needed.

- - - - -
1a98d673 by Sebastian Graf at 2023-08-09T16:24:29-04:00
Cleanup a TODO introduced in 1f94e0f7

The change must have slipped through review of !4412

- - - - -
2274abc8 by Sebastian Graf at 2023-08-09T16:24:29-04:00
More explicit strictness in GHC.Real

- - - - -
ce8aa54c by Sebastian Graf at 2023-08-09T16:24:30-04:00
exprIsTrivial: Factor out shared implementation

The duplication between `exprIsTrivial` and `getIdFromTrivialExpr_maybe` has
been bugging me for a long time.

This patch introduces an inlinable worker function `trivial_expr_fold` acting
as the single, shared decision procedure of triviality. It "returns" a
Church-encoded `Maybe (Maybe Id)`, so when it is inlined, it fuses to similar
code as before.
(Better code, even, in the case of `getIdFromTrivialExpr` which presently
allocates a `Just` constructor that cancels away after this patch.)

- - - - -
d004a36d by Sebastian Graf at 2023-08-09T16:24:30-04:00
Simplify: Simplification of arguments in a single function

The Simplifier had a function `simplArg` that wasn't called in `rebuildCall`,
which seems to be the main way to simplify args. Hence I consolidated the code
path to call `simplArg`, too, renaming to `simplLazyArg`.

- - - - -
8c73505e by Sebastian Graf at 2023-08-09T16:24:30-04:00
Core.Ppr: Omit case binder for empty case alternatives

A minor improvement to pretty-printing

- - - - -
d8d993f1 by Sebastian Graf at 2023-08-09T16:24:30-04:00
Disable tests RepPolyWrappedVar2 and RepPolyUnsafeCoerce1 in JS backend

... because those coerce between incompatible/unknown PrimReps.

- - - - -
f06e87e4 by Sebastian Graf at 2023-08-09T16:24:30-04:00
Inlining literals into boring contexts is OK

- - - - -
4a6b7c87 by Sebastian Graf at 2023-08-09T16:24:30-04:00
Clarify floating of unsafeEqualityProofs (#23754)

- - - - -
b0f4752e by Sebastian Graf at 2023-08-09T16:24:30-04:00
Kill SetLevel.notWorthFloating.is_triv (#23270)

We have had it since b84ba676034, when it operated on annotated expressions.
Nowadays it operates on vanilla `CoreExpr` though, so we should just call
`exprIsTrivial`; thus handling empty cases and string literals correctly.

- - - - -
7e0c8b3b by Sebastian Graf at 2023-08-09T16:24:30-04:00
ANFise string literal arguments (#23270)

This instates the invariant that a trivial CoreExpr translates to an atomic
StgExpr. Nice.

Annoyingly, in -O0 we sometimes generate
```
foo = case "blah"# of sat { __DEFAULT -> unpackCString# sat }
```
which makes it a bit harder to spot that we can emit a standard
`stg_unpack_cstring` thunk.

Fixes #23270.

- - - - -
357f2738 by Sebastian Graf at 2023-08-09T16:24:30-04:00
Deactivate -fcatch-nonexhaustive-cases in ghc-bignum (#23345)

- - - - -
59202c80 by Sebastian Graf at 2023-08-09T16:24:30-04:00
CorePrep: Eliminate EmptyCase and unsafeEqualityProof in CoreToStg instead

We eliminate EmptyCase by way of `coreToStg (Case e _ _ []) = coreToStg e` now.
The main reason is that it plays far better in conjunction with eta expansion
(as we aim to do for arguments in CorePrep, #23083), because we can discard
any arguments, `(case e of {}) eta == case e of {}`, whereas in `(e |> co) eta`
it's impossible to discard the argument.

We do also give the same treatment to unsafeCoerce proofs and treat them as
trivial iff their RHS is trivial.

It is also both much simpler to describe than the previous mechanism of emitting
an unsafe coercion and simpler to implement, removing quite a bit of commentary
and `CorePrepProv`.

In the ghc/alloc perf test `LargeRecord`, we introduce an additional Simplifier
iteration due to #17910. E.g., FloatOut produces a binding
```
lvl_s6uK [Occ=Once1] :: GHC.Types.Int
[LclId]
lvl_s6uK = GHC.Types.I# 2#

lvl_s6uL [Occ=Once1] :: GHC.Types.Any
[LclId]
lvl_s6uL
  = case Unsafe.Coerce.unsafeEqualityProof ... of
    { Unsafe.Coerce.UnsafeRefl v2_i6tr -> lvl_s6uK `cast` (... v2_i6tr ...)
    }
```
That occurs once and hence is pre-inlined unconditionally in the next Simplifier
pass. It's non-trivial to find a way around that, but not really harmful
otherwise. Hence we accept a 1.2% increase on some architectures.

Metric Increase:
    LargeRecord

- - - - -
00d31188 by Sebastian Graf at 2023-08-09T16:24:30-04:00
CorePrep: Eta expand arguments (#23083)

Previously, we'd only eta expand let bindings and lambdas,
now we'll also eta expand arguments such as in T23083:
```hs
g f h = f (h `seq` (h $))
```
Unless `-fpedantic-bottoms` is set, we'll now transform to
```hs
g f h = f (\eta -> h eta)
```
in CorePrep.

See the new `Note [Eta expansion of arguments in CorePrep]` for the details.

We only do this optimisation with -O2 because we saw 2-3% ghc/alloc regressions
in T4801 and T5321FD.

Fixes #23083.

- - - - -
bf885d7a by Matthew Craven at 2023-08-09T16:25:07-04:00
Bump bytestring submodule to 0.11.5, again

Fixes #23789.

The bytestring commit used here is unreleased;
a release can be made when necessary.

- - - - -
7acbf0fd by Sven Tennie at 2023-08-10T19:17:11-04:00
Serialize CmmRetInfo in .rodata

The handling of case was missing.

- - - - -
0c3136f2 by Sven Tennie at 2023-08-10T19:17:11-04:00
Reference StgRetFun payload by its struct field address

This is easier to grasp than relative pointer offsets.

- - - - -
f68ff313 by Sven Tennie at 2023-08-10T19:17:11-04:00
Better variable name: u -> frame

The 'u' was likely introduced by copy'n'paste.

- - - - -
0131bb7f by Sven Tennie at 2023-08-10T19:17:11-04:00
Make checkSTACK() public

Such that it can also be used in tests.

- - - - -
7b6e1e53 by Sven Tennie at 2023-08-10T19:17:11-04:00
Publish stack related fields in DerivedConstants.h

These will be used in ghc-heap to decode these parts of the stack.

- - - - -
907ed054 by Sven Tennie at 2023-08-10T19:17:11-04:00
ghc-heap: Decode StgStack and its stack frames

Previously, ghc-heap could only decode heap closures.

The approach is explained in detail in note
[Decoding the stack].

- - - - -
6beb6ac2 by Sven Tennie at 2023-08-10T19:17:11-04:00
Remove RetFunType from RetFun stack frame representation

It's a technical detail. The single usage is replaced by a predicate.

- - - - -
006bb4f3 by Sven Tennie at 2023-08-10T19:17:11-04:00
Better parameter name

The call-site uses the term "offset", too.

- - - - -
d4c2c1af by Sven Tennie at 2023-08-10T19:17:11-04:00
Make closure boxing pure

There seems to be no need to do something complicated. However, the
strictness of the closure pointer matters, otherwise a thunk gets
decoded.

- - - - -
8d8426c9 by Sven Tennie at 2023-08-10T19:17:11-04:00
Document entertainGC in test

It wasn't obvious why it's there and what its role is.

Also, increase the "entertainment level" a bit.

I checked in STG and Cmm dumps that this really generates closures (and
is not e.g. constant folded away.)

- - - - -
cc52c358 by Finley McIlwaine at 2023-08-10T19:17:47-04:00
Add -dipe-stats flag

This is useful for seeing which info tables have information.

- - - - -
261c4acb by Finley McIlwaine at 2023-08-10T19:17:47-04:00
Add -finfo-table-map-with-fallback -finfo-table-map-with-stack

The -fno-info-table-map-with-stack flag omits STACK info tables from the info
table map, and the -fno-info-table-map-with-fallback flag omits info tables
with defaulted source locations from the map. In a test on the Agda codebase
the build results were about 7% smaller when both of those types of tables
were omitted.

Adds a test that verifies that passing each combination of these flags
results in the correct output for -dipe-stats, which is disabled for the js
backend since profiling is not implemented.

This commit also refactors a lot of the logic around extracting info tables
from the Cmm results and building the info table map.

This commit also fixes some issues in the users guide rst source to fix
warnings that were noticed while debugging the documentation for these flags.

Fixes #23702

- - - - -
d7047e0d by Jaro Reinders at 2023-08-14T04:41:42-04:00
Add changelog entry for specialised Enum Int64/Word64 instances

- - - - -
52f5e8fb by cydparser at 2023-08-14T04:42:20-04:00
Fix -ddump-to-file and -ddump-timings interaction (#20316)

- - - - -
1274c5d6 by cydparser at 2023-08-14T04:42:20-04:00
Update release notes (#20316)

- - - - -
aac87889 by Finley McIlwaine at 2023-08-14T08:34:58-06:00
Refactor estimation of stack info table provenance

This commit greatly refactors the way we compute estimated provenance for stack
info tables. Previously, this process was done using an entirely separate traversal
of the whole Cmm code stream to build the map from info tables to source locations.
The separate traversal is now fused with the Cmm code generation pipeline in
GHC.Driver.Main.

This results in very significant code generation speed ups when -finfo-table-map is
enabled. In testing, this patch reduces code generation times by almost 30% with
-finfo-table-map and -O0, and 60% with -finfo-table-map and -O1 or -O2 .

Fixes #23103

- - - - -


30 changed files:

- .gitlab-ci.yml
- compiler/GHC/Cmm.hs
- compiler/GHC/Cmm/CLabel.hs
- compiler/GHC/Core.hs
- compiler/GHC/Core/Coercion.hs
- compiler/GHC/Core/Coercion/Opt.hs
- compiler/GHC/Core/FVs.hs
- compiler/GHC/Core/Lint.hs
- compiler/GHC/Core/Opt/Arity.hs
- compiler/GHC/Core/Opt/SetLevels.hs
- compiler/GHC/Core/Opt/Simplify/Iteration.hs
- compiler/GHC/Core/Opt/Simplify/Utils.hs
- compiler/GHC/Core/Ppr.hs
- compiler/GHC/Core/TyCo/FVs.hs
- compiler/GHC/Core/TyCo/Rep.hs
- compiler/GHC/Core/TyCo/Subst.hs
- compiler/GHC/Core/TyCo/Tidy.hs
- compiler/GHC/Core/Type.hs
- compiler/GHC/Core/Unfold.hs
- compiler/GHC/Core/Utils.hs
- compiler/GHC/CoreToIface.hs
- compiler/GHC/CoreToStg.hs
- compiler/GHC/CoreToStg/Prep.hs
- compiler/GHC/Driver/Config/CoreToStg/Prep.hs
- compiler/GHC/Driver/Config/StgToCmm.hs
- compiler/GHC/Driver/DynFlags.hs
- compiler/GHC/Driver/Errors/Ppr.hs
- compiler/GHC/Driver/Flags.hs
- compiler/GHC/Driver/GenerateCgIPEStub.hs
- compiler/GHC/Driver/Main.hs


The diff was not included because it is too large.


View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/b196666f0dd1b3d73b0b5a863a1fc911d074c65d...aac878896f5197275d040b3980a57e7464622cf4

-- 
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/compare/b196666f0dd1b3d73b0b5a863a1fc911d074c65d...aac878896f5197275d040b3980a57e7464622cf4
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/20230814/f87c3881/attachment-0001.html>

More information about the ghc-commits mailing list