| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
| |
Recent FreeBSD versions gained the sched_getaffinity function, which made two
mutually exclusive #ifdef blocks to be enabled.
|
| |
|
|
|
|
|
| |
In 35bea01b xxhash.c was removed. Remove the extra-source-files
stanza referring to it.
|
|
|
|
|
| |
Also derive some more instances. GHC doesn't need them, but downstream
consumers may need to e.g. put stuff in maps.
|
|
|
|
|
|
| |
* Move `PRINTF` macro from `Stats.h` to `Stats.c` as it's only needed in
the latter.
* Undefine `PRINTF` at the end of `Messages.h` to avoid leaking it.
|
|
|
|
| |
Fixes #20992.
|
| |
|
|
|
|
| |
Not entirely convinced that this is worth doing.
|
|
|
|
|
|
|
| |
This adds logic, enabled in the `-debug` RTS for checking the internal
consistency of the m32 allocator. This area has always made me a bit
nervous so this should help me sleep better at night in exchange for
very little overhead.
|
|
|
|
|
| |
Renamed to mprotectForLinker and allowed setting of arbitrary protection
modes.
|
|
|
|
|
|
|
|
|
|
|
| |
Previously m32 would assume that the program image was located near the
start of the address space and therefore assume that it wanted pages
in the bottom 4GB of address space. Instead we now check whether they
are within 4GB of whereever the program is loaded.
This is necessary on Windows, which now tends to place the image in high
memory. The eventual goal is to use m32 to allocate memory for linker
sections on Windows.
|
|
|
|
|
|
|
|
|
|
|
| |
Note [Tidying multiple names at once] indicates that if multiple
variables have the same name then we shouldn't prioritise one of them
and instead rename them all to a1, a2, a3... etc
This patch implements that change, some error message changes as
expected.
Closes #20932
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch fixes #17469, by improving matters when you use
non-existent field names in a record construction:
data T = MkT { x :: Int }
f v = MkT { y = 3 }
The check is now made in the renamer, in GHC.Rename.Env.lookupRecFieldOcc.
That in turn led to a spurious error in T9975a, which is fixed by
making GHC.Rename.Names.extendGlobalRdrEnvRn fail fast if it finds
duplicate bindings. See Note [Fail fast on duplicate definitions]
in that module for more details.
This patch was originated and worked on by Alex D (@nineonine)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The commit
commit 18df4013f6eaee0e1de8ebd533f7e96c4ee0ff04
Date: Sat Jan 22 01:12:30 2022 +0000
Define and use restoreLclEnv
omitted to change one setLclEnv to restoreLclEnv, namely
the one in GHC.Tc.Errors.warnRedundantConstraints.
This new commit fixes the omission.
|
|
|
|
| |
c.f. #20980
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, directly calling a function that pattern matches on an
unlifted data type which has at least two constructors in GHCi resulted
in a segfault.
This happened due to unaccounted return frame info table pointer. The fix is
to pop the above mentioned frame info table pointer when unlifted things are
returned. See Note [Popping return frame for unlifted things]
authors: bgamari, nineonine
|
|
|
|
|
|
| |
Previously the documentation was subtly incorrect regarding the bounds
of the operation. Fix this and add a test asserting that a zero-length
operation is in fact a no-op.
|
|
|
|
|
| |
This allows us to produce valid code for indexWord8ArrayAs*# on
platforms that lack unaligned memory access.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
Related CLC proposal is here: https://github.com/haskell/core-libraries-committee/issues/30
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The main motivation for this patch is to allow tests to be added to the
testsuite which test things about the source tree without needing to
build GHC. In particular the notes linter can easily start failing and
by integrating it into the testsuite the process of observing these
changes is caught by normal validation procedures rather than having to
run the linter specially.
With this patch I can run
```
./hadrian/build test --flavour=devel2 --only="uniques"
```
In a clean tree to run the checkUniques linter without having to build
GHC.
Fixes #21029
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Here we introduce a new data structure, RoughMap, inspired by the
previous `RoughTc` matching mechanism for checking instance matches.
This allows [Fam]InstEnv to be implemented as a trie indexed by these
RoughTc signatures, reducing the complexity of instance lookup and
FamInstEnv merging (done during the family instance conflict test)
from O(n) to O(log n).
The critical performance improvement currently realised by this patch is
in instance matching. In particular the RoughMap mechanism allows us to
discount many potential instances which will never match for constraints
involving type variables (see Note [Matching a RoughMap]). In realistic
code bases matchInstEnv was accounting for 50% of typechecker time due
to redundant work checking instances when simplifying instance contexts
when deriving instances. With this patch the cost is significantly
reduced.
The larger constants in InstEnv creation do mean that a few small
tests regress in allocations slightly. However, the runtime of T19703 is
reduced by a factor of 4. Moreover, the compilation time of the Cabal
library is slightly improved.
A couple of test cases are included which demonstrate significant
improvements in compile time with this patch.
This unfortunately does not fix the testcase provided in #19703 but does
fix #20933
-------------------------
Metric Decrease:
T12425
Metric Increase:
T13719
T9872a
T9872d
hard_hole_fits
-------------------------
Co-authored-by: Matthew Pickering <matthewtpickering@gmail.com>
|
|
|
|
| |
Fixes #21002
|
|
|
|
|
|
|
|
|
|
| |
This allows us to remove the dependency on parsec and hence transitively
on text.
Also added some simple unit tests for the parser and fixed two small
issues in the documentation.
Fixes #21033
|
|
|
|
| |
Fixes #14276
|
|
|
|
|
|
| |
Using ghc_plugin_way had the unintended effect of meaning certain tests
weren't run at all when ghc_dynamic=true, if you delete this modifier
then the tests work in both the static and dynamic cases.
|
|
|
|
|
| |
Due to #20791 you need to explicitly flush as otherwise the output from
these tests doesn't make it to stdout.
|
|
|
|
| |
This test was previously not run due to #20960
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
The test produces different output on static vs dynamic GHC builds.
|
| |
|
|
|
|
|
| |
Compare expressions and types when comparing `CmmLoad`s.
Fixes #21016
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch implements a fix for #20817. It ensures that
* The final strictness signature for a function accurately
reflects the unboxing done by the wrapper
See Note [Finalising boxity for demand signatures]
and Note [Finalising boxity for let-bound Ids]
* A much better "layer-at-a-time" implementation of the
budget for how many worker arguments we can have
See Note [Worker argument budget]
Generally this leads to a bit more worker/wrapper generation,
because instead of aborting entirely if the budget is exceeded
(and then lying about boxity), we unbox a bit.
Binary sizes in increase slightly (around 1.8%) because of the increase
in worker/wrapper generation. The big effects are to GHC.Ix,
GHC.Show, GHC.IO.Handle.Internals. If we did a better job of dropping
dead code, this effect might go away.
Some nofib perf improvements:
Program Size Allocs Runtime Elapsed TotalMem
--------------------------------------------------------------------------------
VSD +1.8% -0.5% 0.017 0.017 0.0%
awards +1.8% -0.1% +2.3% +2.3% 0.0%
banner +1.7% -0.2% +0.3% +0.3% 0.0%
bspt +1.8% -0.1% +3.1% +3.1% 0.0%
eliza +1.8% -0.1% +1.2% +1.2% 0.0%
expert +1.7% -0.1% +9.6% +9.6% 0.0%
fannkuch-redux +1.8% -0.4% -9.3% -9.3% 0.0%
kahan +1.8% -0.1% +22.7% +22.7% 0.0%
maillist +1.8% -0.9% +21.2% +21.6% 0.0%
nucleic2 +1.7% -5.1% +7.5% +7.6% 0.0%
pretty +1.8% -0.2% 0.000 0.000 0.0%
reverse-complem +1.8% -2.5% +12.2% +12.2% 0.0%
rfib +1.8% -0.2% +2.5% +2.5% 0.0%
scc +1.8% -0.4% 0.000 0.000 0.0%
simple +1.7% -1.3% +17.0% +17.0% +7.4%
spectral-norm +1.8% -0.1% +6.8% +6.7% 0.0%
sphere +1.7% -2.0% +13.3% +13.3% 0.0%
tak +1.8% -0.2% +3.3% +3.3% 0.0%
x2n1 +1.8% -0.4% +8.1% +8.1% 0.0%
--------------------------------------------------------------------------------
Min +1.1% -5.1% -23.6% -23.6% 0.0%
Max +1.8% +0.0% +36.2% +36.2% +7.4%
Geometric Mean +1.7% -0.1% +6.8% +6.8% +0.1%
Compiler allocations in CI have a geometric mean of +0.1%; many small
decreases but there are three bigger increases (7%), all because we do
more worker/wrapper than before, so there is simply more code to
compile. That's OK.
Perf benchmarks in perf/should_run improve in allocation by a geo mean
of -0.2%, which is good. None get worse. T12996 improves by -5.8%
Metric Decrease:
T12996
Metric Increase:
T18282
T18923
T9630
|
|
|
|
|
|
|
| |
We now get all the commits between the PERF_BASELINE_COMMIT and HEAD and
check any of them for metric changes.
Fixes #20882
|
|
|
|
| |
I also cleaned up the imports slightly while I was at it.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As #20929 pointed out, we were in-elegantly checking for escaping
kinds in `checkValidType`, even though that check was guaranteed
to succeed for type signatures -- it's part of kind-checking a type.
But for /data constructors/ we kind-check the pieces separately,
so we still need the check.
This MR is a pure refactor, moving the test from `checkValidType` to
`checkValidDataCon`.
No new tests; external behaviour doesn't change.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As #20941 describes, this patch implements a couple of small
fixes to the Simplifier. They make a difference principally
with -O0, so few people will notice. But with -O0 they can
reduce the number of Simplifer iterations.
* In occurrence analysis we avoid making x = (a,b) into a loop breaker
because we want to be able to inline x, or (more likely) do
case-elimination. But HEAD does not treat
x = let y = blah in (a,b)
in the same way. We should though, because we are going to float
that y=blah out of the x-binding. A one-line fix in OccurAnal.
* The crucial function exprIsConApp_maybe uses getUnfoldingInRuleMatch
(rightly) but the latter was deeply strange. In HEAD, if
rule-rewriting was off (-O0) we only looked inside stable
unfoldings. Very stupid. The patch simplifies.
* I also noticed that in simplStableUnfolding we were failing to
delete the DFun binders from the usage. So I added that.
Practically zero perf change across the board, except that we get more
compiler allocation in T3064 (which is compiled with -O0). There's a
good reason: we get better code. But there are lots of other small
compiler allocation decreases:
Metrics: compile_time/bytes allocated
---------------------
Baseline
Test Metric value New value Change
-----------------------------------------------------------------
PmSeriesG(normal) ghc/alloc 44,260,817 44,184,920 -0.2%
PmSeriesS(normal) ghc/alloc 52,967,392 52,891,632 -0.1%
PmSeriesT(normal) ghc/alloc 75,498,220 75,421,968 -0.1%
PmSeriesV(normal) ghc/alloc 52,341,849 52,265,768 -0.1%
T10421(normal) ghc/alloc 109,702,291 109,626,024 -0.1%
T10421a(normal) ghc/alloc 76,888,308 76,809,896 -0.1%
T10858(normal) ghc/alloc 125,149,038 125,073,648 -0.1%
T11276(normal) ghc/alloc 94,159,364 94,081,640 -0.1%
T11303b(normal) ghc/alloc 40,230,059 40,154,368 -0.2%
T11822(normal) ghc/alloc 107,424,540 107,346,088 -0.1%
T12150(optasm) ghc/alloc 76,486,339 76,426,152 -0.1%
T12234(optasm) ghc/alloc 55,585,046 55,507,352 -0.1%
T12425(optasm) ghc/alloc 88,343,288 88,265,312 -0.1%
T13035(normal) ghc/alloc 98,919,768 98,845,600 -0.1%
T13253-spj(normal) ghc/alloc 121,002,153 120,851,040 -0.1%
T16190(normal) ghc/alloc 290,313,131 290,074,152 -0.1%
T16875(normal) ghc/alloc 34,756,121 34,681,440 -0.2%
T17836b(normal) ghc/alloc 45,198,100 45,120,288 -0.2%
T17977(normal) ghc/alloc 39,479,952 39,404,112 -0.2%
T17977b(normal) ghc/alloc 37,213,035 37,137,728 -0.2%
T18140(normal) ghc/alloc 79,430,588 79,350,680 -0.1%
T18282(normal) ghc/alloc 128,303,182 128,225,384 -0.1%
T18304(normal) ghc/alloc 84,904,713 84,831,952 -0.1%
T18923(normal) ghc/alloc 66,817,241 66,731,984 -0.1%
T20049(normal) ghc/alloc 86,188,024 86,107,920 -0.1%
T5837(normal) ghc/alloc 35,540,598 35,464,568 -0.2%
T6048(optasm) ghc/alloc 99,812,171 99,736,032 -0.1%
T9198(normal) ghc/alloc 46,380,270 46,304,984 -0.2%
geo. mean -0.0%
Metric Increase:
T3064
|
| |
|
| |
|