Commit the platform definition and build script work necessary to
cross-compile for arm64_32.
arm64_32 is a variant of AARCH64 that supports an ILP32 architecture.
The new _rawHashValue(seed:) requirement allows stdlib types to specialize their hashing when they’re hashed on their own (i.e., not as a component of some composite type).
This makes it possible to get rid of discriminator/terminator values and to eliminate most of Hasher’s resiliency overhead, leading to a measurable speedup, especially for tiny keys.
Beyond switching hashing algorithms, this also enables per-execution hash seeds, fulfilling a long-standing prophecy in Hashable’s documentation.
To reduce the possibility of random test failures, StdlibUnittest’s TestSuite overrides the random hash seed on initialization.
rdar://problem/24109692
rdar://problem/35052153
This is causing the the test to be largely ignored, and disrupts regular testing
(including in continuous integration) where an integration can fail just
because this tests fails for "expected" reasons.
We should re-enable the test once it is has a way to tolerate the
"expected" failure, perhaps using some notion of statistical confidence
to determine if the failure is real.
Bump the number of trials from 10 to 14, which makes the test orders of
magnitude less likely to fail. For a range of size 10, doing 10 trials
meant that a single call to checkRange would fail ~0.03% of the time;
with 14 trials the spurious failure rate is ~0.0003%. We have 10 calls
to checkRange in this test with ranges of size <= 10.
While this test didn't fail that often before, we have a very large
number of automated builds and it has been a constant low-level source
of friction that this test fails and we have to look at and then ignore
the results.
This patch adds powerpc64le Linux support. While the patch also adds
the matching powerpc64 bits, there are endian issues that need to be
sorted out.
The PowerPC LLVM changes for the swift ABI (eg returning three element
non-homogeneous aggregates) are still in the works, but a simple LLVM
fix to allow those aggregates results in swift passing all but 8
test cases.
The standard library has grown significantly, and we need a new
directory structure that clearly reflects the role of the APIs, and
allows future growth.
See stdlib/{public,internal,private}/README.txt for more information.
Swift SVN r25876
The new modules are:
* SwiftUnstable -- assorted additions to the core standard library
(more algorithms etc.)
* SwiftUnstableDarwinExtras -- proposed additions to the Darwin overlay,
not yet reviewed by the Darwin team.
* SwiftUnstablePthreadExtras -- wrappers that make it possible to
use pthread in Swift (they work around the lack of block-based API in
pthread). In future these could be possibly folded into the Darwin
overlay as well.
These APIs are useful without StdlibUnittest for writing automation
tools in Swift. Just like SwiftExperimental, none of these modules are
exposed to extrenal users.
Also, since these new modules can be compiled with -sil-serialize-all
(unlike StdlibUnittest, where we can't apply the flag because of
compiler bugs), standard library tests that need to run optimized code
(like AtomicInt.swift) are *much* faster now.
Swift SVN r25679
Now that it disables access control, this lit definition applies to many
more tests, where we used to use %target-run-stdlib-swift before access
control came online.
Also, drop -Xfrontend -disable-access-control from a bunch of tests that
don't need it.
Swift SVN r22634
* update stale values for hashes (the previous one were from one of my
earlier hash implementations)
* add a separate code path for 32-bit where Int type is used
Swift SVN r21147
This function mixes the bits in the hash value, which improves Dictionary
performance for keys with bad hashes.
PrecommitBenchmark changes with greater than 7% difference:
``````````Dictionary2`,```1456.00`,```1508.00`,```1502.00`,````624.00`,````607.00`,````592.00`,`864.00`,``145.9%
``````````Dictionary3`,```1379.00`,```1439.00`,```1408.00`,````585.00`,````567.00`,````552.00`,`827.00`,``149.8%
````````````Histogram`,````850.00`,````849.00`,````851.00`,```1053.00`,```1049.00`,```1048.00`,`199.00`,``-19.0%
````````````````Prims`,```1999.00`,```2005.00`,```2018.00`,```1734.00`,```1689.00`,```1701.00`,`310.00`,```18.4%
``````````StrSplitter`,```2365.00`,```2334.00`,```2316.00`,```1979.00`,```1997.00`,```2000.00`,`337.00`,```17.0%
```````````````TwoSum`,```1551.00`,```1568.00`,```1556.00`,```1771.00`,```1741.00`,```1716.00`,`165.00`,```-9.6%
Regressions are in benchmarks that use `Int` as dictionary key: we are just
doing more work than previously (hashing an `Int` was an identity function).
rdar://17962402
Swift SVN r21142
