https://pastebin.com/CmnzwUFN
The key idea is to avoid using 16 byte NEON and use 8 byte NEON which has lower latency for BitMask::Match. Even though 16 byte NEON achieves higher throughput, in SwissMap it's very important to catch these Matches with low latency as probing on average happens at most once.
I also introduced NonIterableMask as ARM has really great cbnz instructions and additional AND on scalar mask had 1 extra latency cycle
PiperOrigin-RevId: 453216147
Change-Id: I842c50d323954f8383ae156491232ced55aacb78
Benchmarks: https://pastebin.com/tZ7dr67W. Works well especially on smaller ranges.
After a week on spending optimizing NEON SIMD where I almost managed to make hash tables work with NEON SIMD without performance hits (still 1 cycle to optimize and I gave up a little), I found an interesting optimization for aarch64 to use cls instruction (count leading sign bits).
The loop has a property that ctrl_ group is not matched against count when the first slot is empty or deleted.
```
void skip_empty_or_deleted() {
while (IsEmptyOrDeleted(*ctrl_)) {
uint32_t shift = Group{ctrl_}.CountLeadingEmptyOrDeleted();
ctrl_ += shift;
slot_ += shift;
}
...
}
```
However, `kEmpty` and `kDeleted` have format of `1xxxxxx0` and `~ctrl & (ctrl >> 7)` always sets the lowest bit to 1.
In naive implementation, it does +1 to start counting zero bits, however, in aarch64 we may start counting one bits immediately. This saves 1 cycle and 5% of iteration performance.
Then it becomes hard to find a supported and sustainable C++ version of it.
`__clsll` is not supported by GCC and was supported only since clang 8, `__builtin_clrsb` is not producing optimal codegen for clang. `__rbit` is not supported by GCC and there is no intrinsic to do that, however, in clang we have `__builtin_bitreverse{32,64}`. For now I decided to enable this only for clang, only if they have appropriate builtins.
PiperOrigin-RevId: 451168570
Change-Id: I7e9256a60aecdc88ced4e6eb15ebc257281b6664
Add a new (internal) feature test macro to detect whether the wrappers are no-ops on a given platform.
Note that one-arg __builtin_prefetch(x) is equivalent to __builtin_prefetch(x, 0, 3), per `man BUILTIN_PREFETCH(3)` and gcc docs.
PiperOrigin-RevId: 449508660
Change-Id: I144e750205eec0c956d8dd62bc72e10bdb87c4f7
--
291f7ef542f73e4801ab5108014bc02344ef31df by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 437835981
Change-Id: I42fd92e74903894533ac9984d7f622e3ba20f468
--
2e8caf1a57c50b518e05b4bca48e4fe1bb19af82 by Andy Getzendanner <durandal@google.com>:
Internal change
PiperOrigin-RevId: 437832673
Change-Id: I61b35089418d01a54cecf161b254b68252bebff3
--
b927482ccc399f7e337b60582988b914d9946e4e by Derek Mauro <dmauro@google.com>:
Simplify endian intrinsics for modern compilers
All modern compilers have either __builtin_bswapN (gcc, clang) or
_byteswap_TYPE (MSVC). The other intrinsic definitions are no longer
necessary.
PiperOrigin-RevId: 437772295
Change-Id: Ifb3d88ba24b9097f87ceb202272b36d2f5e5117f
--
b6782a2247a16d5c14706a74ec577c19963d9f97 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 437373174
Change-Id: I0f77e1780dee90d7a3c32a08d96c4aeb624a57b4
--
a53e0c724e37b0b01515a99bd25394b8e21ffdfc by Derek Mauro <dmauro@google.com>:
Unify detection of SSE2 and SSSE3 instruction sets and
include the proper headers
Fix the intrinsic implementation of FastHexToBufferZeroPad16 in
numbers.h which only relies on SSSE3, not SSE 4.2.
https://godbolt.org/z/Pf5bn1Yv9Closes#639
PiperOrigin-RevId: 437286940
Change-Id: Ic97948399b61b91e9c0bccd09313b795b904d714
--
f173f597cb2a75ef2a989f45a496334b85e6f40d by Abseil Team <absl-team@google.com>:
Change assertion function to enable clearer error messages.
PiperOrigin-RevId: 437227057
Change-Id: If420d2f63b51feef6648762f344d5be012cd9c85
GitOrigin-RevId: 291f7ef542f73e4801ab5108014bc02344ef31df
--
7fbe730c56c0890f27df47f9ad4f0cfa600c6ad9 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 424639048
--
1825fcf704e45e881cdd1d2fe3eb2226f523f3cd by Derek Mauro <dmauro@google.com>:
Disable tests Table.EnsureNonQuadraticTopNXorSeedByProbeSeqLength and Table.EnsureNonQuadraticTopNLinearTransformByProbeSeqLength again
as they are (still) flaky
PiperOrigin-RevId: 424616745
--
e624119171d9bceec2f6df023b220a6642070cfb by Abseil Team <absl-team@google.com>:
Switch from *_TEST_CASE_P to *_TEST_SUITE_P gtest macros.
PiperOrigin-RevId: 424404790
--
a110698c10554fb1f640412db8cd7c5a38b7b8e8 by Abseil Team <absl-team@google.com>:
Remove extraneous newline in documentation.
PiperOrigin-RevId: 424328502
GitOrigin-RevId: 7fbe730c56c0890f27df47f9ad4f0cfa600c6ad9
Change-Id: I56fd8c7cecf65fe37cb4cf2dde70842867d637d7
--
75504b9d2eb7560359b98b69136d071f980e04f2 by Abseil Team <absl-team@google.com>:
Fix typos in documentation.
PiperOrigin-RevId: 423376798
--
bf87e5de48a868f49a57d516be027e6e3a3cc3bd by Gennadiy Rozental <rogeeff@google.com>:
Correct WEAK attribute enabling condition.
ABSL_ATTRIBUTE_WEAK is present if:
compiler has built-in attribute weak
OR
we are using gcc (and not clang)
AND
we are not on windows
OR
we use windows clang version >= 9.0.0
AND
we are not on MinGW
PiperOrigin-RevId: 423357629
--
a01a8f1b7ea3da4ec349db452162a3333953dd9d by Abseil Team <absl-team@google.com>:
There are magic numbers in the expected load factors and probe lengths, and they seem to be wrong especially under msvc. Even under the linux tool chain, these tests fail occasionally. Fix the magic numbers to make the tests succeed.
PiperOrigin-RevId: 423320829
--
fd109295a1425ca1cb2b69fe34a294b6189542c0 by Laramie Leavitt <lar@google.com>:
Manually align buffers in randen_engine.
In C++ it's implementation defined whether types with extended alignment are supported.
randen_engine uses vector intrinsics with 16-byte alignment requirements in
some instances, so internally we allocate an extra 8 bytes to manually align to 16.
No detectable performance impact.
PiperOrigin-RevId: 423109265
GitOrigin-RevId: 75504b9d2eb7560359b98b69136d071f980e04f2
Change-Id: I8c5ab2269ff6d9e89d3b4d0e95d36ddb6ce8096e
--
e7f53dfbf809812e84770217777f81b6308a3084 by Abseil Team <absl-team@google.com>:
Add a parameter pack to absl profile to allow profiles to separate
dynamic data from static data that is available at constructor-time.
Background: `inline_element_size` is effectively constant, but there
is a data race between its initialization and its access. We had fixed that race by making
inline_element_size atomic. This CL changes `inline_element_size`
back to a non-atomic integer, and provides a way for all profiles to
provide Register()-time values.
PiperOrigin-RevId: 413960559
--
70234c5943f8e37e17c1d9c54d8ed61d39880abf by Chris Kennelly <ckennelly@google.com>:
Document that absl::FunctionRef does not allocate.
PiperOrigin-RevId: 413946831
--
3308ae571412c4be3cc32d088c6edac98ff2d1ed by Samuel Benzaquen <sbenza@google.com>:
Internal change
PiperOrigin-RevId: 413933619
--
1617093a730d055edcf7bc04fdd6509783f5f75d by Martijn Vels <mvels@google.com>:
Internal Change
PiperOrigin-RevId: 413778735
--
03ad683f059c806a6c8b04f5b79b2662c3df8c73 by Evan Brown <ezb@google.com>:
Unify btree erase_if definitions and optimize them so that we only do rebalancing once per leaf node.
PiperOrigin-RevId: 413757280
--
5ba402f70801938178e486617063f01c7862525d by Martijn Vels <mvels@google.com>:
Cleanup up cord sampling internals
PiperOrigin-RevId: 413755011
--
522da8f9d3e0f11630d89fb41952004742bc335a by Evan Brown <ezb@google.com>:
Add b-tree benchmark for erase_if.
Since this benchmark doesn't work for std:: containers before C++20, disable it for them.
PiperOrigin-RevId: 413740844
--
a690ea42de8ed4a761d00235d8b2fb7548ba9732 by Andy Getzendanner <durandal@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 413735737
GitOrigin-RevId: e7f53dfbf809812e84770217777f81b6308a3084
Change-Id: I4f9f9039ba92831bc48971964aa063244c9fed72
--
08d99ee216b7bfac1c5182db952d4e053e5ebc31 by Abseil Team <absl-team@google.com>:
Fix race condition reported by tsan on `inline_element_size` in hashtablez.
PiperOrigin-RevId: 413520771
GitOrigin-RevId: 08d99ee216b7bfac1c5182db952d4e053e5ebc31
Change-Id: Ibd396803f04a659cfbdb8dc7ec37511643657694
--
05a099a580753f8e96cee38572e94dcdc079361b by Abseil Team <absl-team@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 405966217
--
c6b81e9ebc183d8389f14ecd091c8bad08cfe0aa by Abseil Team <absl-team@google.com>:
Add `inline_element_size` to hashtablez (so that we can compute the weighted load factors properly e.g., in b/187896534).
PiperOrigin-RevId: 405917711
--
3e3673de4e54e4142c54b09e1644dfa3de4bb296 by Abseil Team <absl-team@google.com>:
align indent of code comment in mutex.h
PiperOrigin-RevId: 405871997
--
2248301a5b14f8d2be5b2e9088f3528a353ea491 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 405639236
--
bc7d3c56fdad3dde4b89324af142529f2afe5f1b by Abseil Team <absl-team@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 405508045
--
66472387276ef02505d99195747be862768bb35b by Laramie Leavitt <lar@google.com>:
Also use uint8_t golden values in randen_test.cc
This makes randen_test, randen_slow_test, and randen_hwaes_test essentially
identical, as is the intent.
PiperOrigin-RevId: 405484423
GitOrigin-RevId: 05a099a580753f8e96cee38572e94dcdc079361b
Change-Id: I3dd5b0cfdb98d6e1ab02266194ba67d15428c2f8
--
3794fe8db7a8e5a17945a8d6e198cde1db1fed7d by Chris Kennelly <ckennelly@google.com>:
Record maximum reserve or rehash argument in hashtable statistics.
This makes it possible to identify containers that are large because of reserve
calls or ones that could have an opportunity for more precise capacity sizing.
PiperOrigin-RevId: 396851064
--
c3d247c08acfd45d8e19cfd8e6e841e16e38e23d by Abseil Team <absl-team@google.com>:
Remove extra semi-colon
PiperOrigin-RevId: 396823800
GitOrigin-RevId: 3794fe8db7a8e5a17945a8d6e198cde1db1fed7d
Change-Id: I9f26407c2dc6b3dff04f6f3e249571dd8ab288c3
--
69b5d0b2a5adb49a53e51f9da6848eaa484242fe by Derek Mauro <dmauro@google.com>:
Changes the absl::Duration factory functions to disallow types that
are convertible to int or double, instead requiring that the argument
itself is indeed an integer or a floating-point number. This will
prevent callers from passing arguments, such as std::atomic<T>.
This change is an API break. Information and a tool to fix issues
can be found at
https://abseil.io/docs/cpp/tools/upgrades/duration-conversions
PiperOrigin-RevId: 387153494
--
786063e438ab6a55ac4baa88ad4d20a8293be52a by Evan Brown <ezb@google.com>:
Make ctrl_t be an enum class.
This adds type safety, and also when strict aliasing is enabled, the compiler will know that control bytes can't alias non-control bytes.
Also make H2() return h2_t.
PiperOrigin-RevId: 387120717
--
7e537aabec1c255d6e7c9d21232c597c1c2077bf by Evan Brown <ezb@google.com>:
Add some missing `const` keywords to ctrl_t* function parameters.
PiperOrigin-RevId: 386976062
--
da53ac6d91cabd951e81dd0a145e1e52b918955f by Martijn Vels <mvels@google.com>:
Change Seek and InitOffset to return nullptr instead of assert / fail.
This makes it consistent with the rest of the API (Next, Previous, Skip) and hardens it against invariants that are harder (or less likey) to be upheld correctly by the caller.
PiperOrigin-RevId: 386963283
--
a4d1faac020d5025edf53ce81808e5db68da7d89 by Abseil Team <absl-team@google.com>:
PC / Backtrace / Symbolization for Emscripten.
PiperOrigin-RevId: 386957724
--
97f2c47d83ba9d3ac89e1f55bd06897686ffd063 by Martijn Vels <mvels@google.com>:
Fix static casts ([-Wimplicit-int-conversion])
PiperOrigin-RevId: 386951646
--
9530c795248543817cbc4013953baa09c35f5e1a by Abseil Team <absl-team@google.com>:
Fix incorrect header guard in cord_rep_btree_navigator.h
PiperOrigin-RevId: 386907904
--
90ce5872406df2b7f4c428683741dc13a572267e by Abseil Team <absl-team@google.com>:
Small grammar fixes for some StatusCode descriptions.
PiperOrigin-RevId: 386906217
--
b30a2fd777f12a04a4d512f37a34614b0d05ce99 by Derek Mauro <dmauro@google.com>:
Skip length checking when constructing absl::string_view from std::string.
The length check causes unnecessary code bloat.
PiperOrigin-RevId: 386857974
--
fa171536c359bfa2a1b80297e844519bb9ee7791 by Martijn Vels <mvels@google.com>:
Introduce CordRepBtreeNavigator
CordRepBtreeNavigator implements bi-directional navigation over all data edges stored inside a Cord Btree.
PiperOrigin-RevId: 386519102
GitOrigin-RevId: 69b5d0b2a5adb49a53e51f9da6848eaa484242fe
Change-Id: I1b35188d66133f8cb73d346bc5564aac4e0b3e80
--
2684e80d877b688b8d9e0af1b7acddbadc973152 by Evan Brown <ezb@google.com>:
Add an insert codegen function for raw_hash_set_benchmark.
PiperOrigin-RevId: 381052237
--
8394ef3071714a41484cb5b271cba0611d954a7a by Abseil Team <absl-team@google.com>:
Optimize raw_hash_set ctor for random-access iterators
PiperOrigin-RevId: 380832215
GitOrigin-RevId: 2684e80d877b688b8d9e0af1b7acddbadc973152
Change-Id: Icf7929fdfab50a1b26f3dc5505575363b4f5838d
--
b1fc72630aaa81c8395c3b22ba267d938fe29a2e by Derek Mauro <dmauro@google.com>:
Fix -Wdeprecated-copy warnings from Clang 13.
Example:
error: definition of implicit copy assignment operator for 'UDT' is deprecated because it has a user-declared copy constructor [-Werror,-Wdeprecated-copy]
PiperOrigin-RevId: 380058303
--
0422744812b1a2010d9eea5b17fbe89f3441b66b by Evan Brown <ezb@google.com>:
Change the "full table!" asserts in raw_hash_set to use `<= capacity` instead of `< capacity`.
If we add support for non-power-of-two-minus-one capacities, this is the correct thing to assert. For example, consider: Group::kWidth = 8, capacity_ = 8, ctrl_ = {kEmpty, 1, 2, 3, 4, 5, 6, 7, kSentinel, kEmpty, 1, 2, 3, 4, 5, 6}. In this case, if we do an unsuccessful lookup with H2 mapping to slot 1, then the first Group will contain {1, 2, 3, 4, 5, 6, 7, kSentinel} so we need to continue to the second Group (at which point seq.index() == 8 == capacity_) to find a kEmpty.
Note: this is a no-op change for now since we never have `capacity % Group::kWidth == 0`.
PiperOrigin-RevId: 380033480
--
40628c34d540356de65fabb16c1439c0ec7a0764 by Abseil Team <absl-team@google.com>:
Drop out-of-date documentation about `absl::FixedArray`'s allocator support
PiperOrigin-RevId: 379811653
--
e7ad047863ae55c9b7aec0753cfc527a4ea614bc by Evan Brown <ezb@google.com>:
Fix a bug in ConvertDeletedToEmptyAndFullToDeleted in which we were copying 1 more cloned control byte than actually exists.
When alignof(slot_type)>1, this wouldn't cause a problem because the extra byte is padding.
Also change loop bounds to not rely on the fact that capacity_+1 is a multiple of Group::kWidth.
PiperOrigin-RevId: 379311830
--
1a3ba500fb2c33205854eb9258cd6e0fb1061bca by Martijn Vels <mvels@google.com>:
Change Ring, EXTERNAL and FLAT tag values to be consecutive values
The purpose of this change is to have FLAT = EXTERNAL + 1. Especially in the ring and btree alternative code, there is a common check if a node is a 'plain' edge (EXTERNAL or FLAT), or 'something else'. This change can make that check a single branch, i.e., instead of 'tag == EXTERNAL || tag >= FLAT', we can simply check for 'tag >= EXTERNAL'. Likewise we have some cases where we check for RING, EXTERNAL or FLAT, so we align RING + 1 with EXTERNAL.
PiperOrigin-RevId: 379291576
--
0c78e65ca4d85244b106c3f8e24cf268e09e72a3 by Benjamin Barenblat <bbaren@google.com>:
Round a double multiplication before casting it to integer
The code
static_cast<int>(x * y)
(for double x and y) performs a double multiplication into a temporary
that, by standard, may have excess precision. The subsequent cast to int
discards the excess precision. However, the cast may examine the excess
precision during conversion, producing surprising results like
static_cast<int>(1.7 * 10) == 16
on certain systems. Correct this case by explicitly rounding 1.7 * 10
before casting it.
PiperOrigin-RevId: 378922064
GitOrigin-RevId: b1fc72630aaa81c8395c3b22ba267d938fe29a2e
Change-Id: Ica708a006921118673e78d5fd2d61fe0fb0894d1
--
066144400e12616f6771e512427bcd98aa203455 by Abseil Team <absl-team@google.com>:
Internal comment cleanup.
PiperOrigin-RevId: 377368470
--
3ba56d263bd90a9797d12b5ce29edce3fa65917c by Abseil Team <absl-team@google.com>:
Add tests for hash table capacity being unchanged by insert.
PiperOrigin-RevId: 377303140
GitOrigin-RevId: 066144400e12616f6771e512427bcd98aa203455
Change-Id: I2edf14b412e45a2ad490dcf9f06e009c12a60e3e
--
8e75347c10d85112296811be6ef35761744ad9bc by Derek Mauro <dmauro@google.com>:
Big update to LTS release process
* Add create_lts.py script to to the LTS modification
This is simpler than copybara since very few changes are needed
* Use the default installation paths instead of a versioned path.
If a versioned path is needed, this is easy to change on the commandline.
* Make the integration test use the LTS transformed version
* Test both static and dynamic linking (fixes pkg-config dynamic linking)
PiperOrigin-RevId: 363566934
--
e00e971a2de3138861f5e1900201c9cc7788f714 by Laramie Leavitt <lar@google.com>:
Add a non-compile test to absl::BitGenRef for temporaries.
PiperOrigin-RevId: 363437284
--
3685644ec115d99789de32aceb76c32a00756fea by Derek Mauro <dmauro@google.com>:
Make OSS code consistent with internal code by using the forward
declaration of absl::Status that contains ABSL_MUST_USE_RESULT.
PiperOrigin-RevId: 363426906
--
b85fec142c3aa3f632fa985f9f8f73a253819723 by Evan Brown <ezb@google.com>:
Move raw_hash_set::infoz_ into raw_hash_set::settings_. This reduces the size of raw_hash_sets by alignof(size_t) bytes when hashtablez is disabled.
PiperOrigin-RevId: 363034264
--
c6fde3b17e5845191eb8b2bfc1760c8bfb9573ff by Mark Barolak <mbar@google.com>:
Internal change
PiperOrigin-RevId: 362990378
--
81713cf964905b43d1cbe32ce5fed97539029625 by Abseil Team <absl-team@google.com>:
Fix typo in comment (execeptions -> exceptions).
PiperOrigin-RevId: 362946191
--
3ee92ca470feca44da417b03ee45a915c6eb5155 by Abseil Team <absl-team@google.com>:
Add absl::FindAndReportLeaks and routes it to the corresponding
__lsan_do_recoverable_leak_check.
PiperOrigin-RevId: 362622199
--
b95b7194b20e02c20d72289fbc79a0d35b82e256 by Abseil Team <absl-team@google.com>:
Add `kWithEverything` to StatusToStringMode
PiperOrigin-RevId: 362595218
--
0a960d96a0014eab7e1c55b479269450ed8e98d7 by Abseil Team <absl-team@google.com>:
Accept e.g. ".__uniq" as a valid clone name. Further, bring the implementation
on par with libiberty's demangler grammar.
Clang introduced option -funique-internal-linkage-names that adds the suffix
".__uniq.[0-9]+" to internal linkage functions to give them a globally unique
identifier. The suffix was designed to work with existing demanglers which do
recognize a "_" along with the alphanumeric string. This change enhances the
demangler to allow "_" with the alphanumeric string.
Please refer to libiberty's cp-demangle.c where function d_clone_suffix
implements the demangling of clone suffixes :
1. '_' is accepted as a valid character with the alphanumeric sequence.
2. The alphanumberic sequence is optional.
3. The digit sequence is optional.
PiperOrigin-RevId: 362557420
--
2ac5ea212c150afd2f58025a5cab8c45d16949c6 by Abseil Team <absl-team@google.com>:
Change variable name 'slots' to 'slot_count' to avoid name-clash with Qt builds.
PiperOrigin-RevId: 362556289
--
934f0f409c9c548716a46363d6e243406fad4028 by Mark Barolak <mbar@google.com>:
Clarify the comment on ABSL_CACHELINE_SIZE to indicate that the macro definition itself shouldn't change, but rather that call sites should change when possible.
This addresses the request for improved documentation in https://github.com/abseil/abseil-cpp/pull/842.
PiperOrigin-RevId: 362354288
GitOrigin-RevId: 8e75347c10d85112296811be6ef35761744ad9bc
Change-Id: I33ec8561d8d645c3353e9d2dd447501d0e1825a7
--
f9476c95cf7625d7b0fc4661f253b0aac4341044 by Abseil Team <absl-team@google.com>:
Add a test to verify that the new checksum field in Hashtablez is calculated
PiperOrigin-RevId: 356744293
--
ff8a3612463000e8c3d451e50367a3c65cb6cf21 by Abseil Team <absl-team@google.com>:
Remove the implied support comment for port.h, attributes.h, and integral_types.h's C compatibility from the header documentations.
Abseil-cpp is a C++ library; this brings port.h, attributes.h, and integral_types.h, into our stance for the rest of Abseil (aka, no assurance of C compatibility)
There is no guarantee that future changes to port.h, attributes.h, and integral_types.h, and their dependencies, will remain compatible with C, even for macros and definitions that currently are.
PiperOrigin-RevId: 356727505
--
be62292016381deee628dbb3f36cb6009bcc0282 by Abseil Team <absl-team@google.com>:
internal change
PiperOrigin-RevId: 356608125
--
13b35f17171df3d6853ea7088797b3be611505fc by Evan Brown <ezb@google.com>:
Clarify the comments for CapacityToGrowth/GrowthToLowerboundCapacity methods to specify the intent that capacity should equal growth when `capacity+1 < kWidth`.
Also add testing for this behavior.
PiperOrigin-RevId: 356579041
GitOrigin-RevId: f9476c95cf7625d7b0fc4661f253b0aac4341044
Change-Id: Iadd094d109b4869998f2427319ef66d1cf1e8eff
--
6e9f93888bbe6718997ed90bbd049f1f3572b066 by Abseil Team <absl-team@google.com>:
Fix status to be safe for self move assignment.
PiperOrigin-RevId: 346815392
--
35cae74a977f258e81dfbe925fb5a34cb6632036 by Gennadiy Rozental <rogeeff@google.com>:
Eliminate unnecessary access to the global vars.
PiperOrigin-RevId: 346777168
--
e7e786c243069060f5d6c1c945decb4b0b83f95b by Andy Getzendanner <durandal@google.com>:
Internal change.
PiperOrigin-RevId: 346685656
--
4ccd41c48f1a83cfa20b3ea534f743dd7d788376 by Abseil Team <absl-team@google.com>:
Move CordRep Ref() and Unref() logic into cord_internal.h
This change moves Ref() and Unref() logic out of cord.cc into cord_internal. The main purpose is to make upcoming ring buffer changes easier to isolate from existing cord.cc code. Notice that this removes the nullptr check from Unref() and now requires it to be non null (which held true most times). We may need to rethink if the 'unref unlikely one' is the common case: are cordreps most likely shared? This may be something between 'root' and non root nodes, i.e., is it more likely for leaf / flat nodes in large cords to be shared than top level cordreps being shared? Vice versa? Benchmarks say that we mostly shouldn't care, the caveat being that atomic ops seem more expensive on upcoming archs (arcadia) so we should error on the side of an extra IsOne() branch saving us single owned atomic ops.
PiperOrigin-RevId: 346676121
--
f0babab103b9e60d61ba09482d468985e43eceb3 by Samuel Benzaquen <sbenza@google.com>:
Fix iterator based constructor and `.insert` members to only require
EmplaceConstructible as the standard specifies.
PiperOrigin-RevId: 346616707
--
8f48eedda02277f9c96a88ed7726e34b557cce20 by Evan Brown <ezb@google.com>:
Fix a bug in binary_search_impl when there's a transparent, three-way comparator that has different equivalence classes for different lookup types.
Add a new can_have_multiple_equivalent_keys method to share the common logic for these cases.
PiperOrigin-RevId: 346605948
--
649183cb3cc9383431de9c81fb1c0f885d4001ae by Abseil Team <absl-team@google.com>:
Add benchmark for accessing a Duration flag.
PiperOrigin-RevId: 346594843
--
fefdb046520871af63ce2229e2f7cccfc0483dea by Abseil Team <absl-team@google.com>:
Restructure CordReader for upcoming ring buffer changes.
PiperOrigin-RevId: 346410642
--
8b2f50e7da0ebab06ead5f94e366e984ca23cb6a by Abseil Team <absl-team@google.com>:
Wire in an internal-only flag to toggle upcoming ring buffer changes on/off for experimentation.
PiperOrigin-RevId: 346199111
GitOrigin-RevId: 6e9f93888bbe6718997ed90bbd049f1f3572b066
Change-Id: I8f34866b25a79209cb5448bbb28dd3044111d2e9
--
4ee535c37f92cd45b8c9aa009e5c833265b3a0bb by Samuel Benzaquen <sbenza@google.com>:
Test and fix `insert(hint, node)`
PiperOrigin-RevId: 337122891
--
7b760ced555756fecbad702fedb697424dd65167 by Abseil Team <absl-team@google.com>:
Changed the minimum version of iOS and OSX for Abseil Podspec.
PiperOrigin-RevId: 336926756
GitOrigin-RevId: 4ee535c37f92cd45b8c9aa009e5c833265b3a0bb
Change-Id: I94e70f3342570c83b9965ca458a3f02eaa3efc0d
--
a5af5874c1c5cc02bd2a748d455321f82b6f2a93 by Andy Getzendanner <durandal@google.com>:
fix compile fails with asan and -Wredundant-decls
Import of https://github.com/abseil/abseil-cpp/pull/801
PiperOrigin-RevId: 336693223
--
ed9df42ab2b742386c6692c2bed015374c919d9c by Derek Mauro <dmauro@google.com>:
Fix integer conversion warning
Fixes#814
PiperOrigin-RevId: 336651814
--
0ab4c23884e72dce17b67c1eb520f9dbb802565d by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 336585378
--
eba0e3dccd52a6e91bcff84075bef0affc650b74 by Matt Kulukundis <kfm@google.com>:
Add bitset operations to Futex helper.
PiperOrigin-RevId: 336409368
--
8b0709a8b4500bf5f0af4b602d76a298d81645e8 by Abseil Team <absl-team@google.com>:
Fix code indentation in a comment.
PiperOrigin-RevId: 336368167
--
bc3961c87a7e7760c10319a5b0349c279f7ae3ad by Samuel Benzaquen <sbenza@google.com>:
Improve performance of the registry:
- Reduce contention
- Reduce memory usage for each flag by `6*sizeof(void*)`.
- Replace one immortal allocation per-flag with a single one for all the flags
- Slightly improve single-threaded performance by avoiding the std::map indirections.
PiperOrigin-RevId: 336365904
--
264ad9f28f935aad8b6b1437f8bf804fa9104346 by Abseil Team <absl-team@google.com>:
Fix typo in comment on absl::Condition.
PiperOrigin-RevId: 336311680
--
b5b808a8c75ca0df7b09eff9a423ec171d80f771 by Derek Mauro <dmauro@google.com>:
Add missing Apache license headers
PiperOrigin-RevId: 336294980
--
89446c3a4793df8b95060385cf3e219357c3db1d by Andy Soffer <asoffer@google.com>:
Internal changes
PiperOrigin-RevId: 336287465
--
57c8be4e294881bc79a6a44b8e4bf7ecbb19b9b9 by Matt Kulukundis <kfm@google.com>:
Extract Futex from an implementation detail of Wait to a private interface.
PiperOrigin-RevId: 336123209
GitOrigin-RevId: a5af5874c1c5cc02bd2a748d455321f82b6f2a93
Change-Id: Ie5a0ebe28e571814e3e11d4c05ca308523ccf311
--
0e6d5abe305df4e943e0bc0256c67afc00956691 by Abseil Team <absl-team@google.com>:
Ensure that Hashtablez does not pull in absl::Mutex dependency when turned off at compile time.
PiperOrigin-RevId: 331142625
GitOrigin-RevId: 0e6d5abe305df4e943e0bc0256c67afc00956691
Change-Id: I984d6d3436e4825a7b20758c45b3142c0f3bb45a
--
587e6db882749fa7faa12815e614afab04d218b9 by Derek Mauro <dmauro@google.com>:
Use attribute detection for other sanitizer related attributes
PiperOrigin-RevId: 324077073
--
3ee55e4935b4235516b1fcac3c55945e510f7afc by Evan Brown <ezb@google.com>:
Simplify CordRepExternal allocation/deallocation.
I think this can save some memory when `Releaser` is empty and when on platforms where alignof(CordRepExternal) < (default `::operator new` alignment).
We no longer need the API requirement that alignof(Releaser) <= (default `::operator new` alignment).
Also remove another static_assert from a TODO in cord_internal.h and fix some warnings about calling std::move on a forwarding reference.
PiperOrigin-RevId: 324053720
--
9fc78436565eb3b204d4aa425ee3773354392f45 by Derek Mauro <dmauro@google.com>:
Use auto-detected sanitizer attributes for ASAN, MSAN, and TSAN builds
PiperOrigin-RevId: 323831461
GitOrigin-RevId: 587e6db882749fa7faa12815e614afab04d218b9
Change-Id: Ie0e4a2846d7f66988a2d81a5e50721b62fdb3d6d
--
1f80d4f1cb8847545627a2d0b18f697c4a763292 by Samuel Benzaquen <sbenza@google.com>:
Add a hint message to the assertions for easier debugging.
Also, move it out of the template to avoid printing the whole template
instantiation as part of the assertion. It is not relevant and can
significantly bloat the error message and hide the important bits.
PiperOrigin-RevId: 316736140
--
223e97a90150de5b7206be2e45c62a9b70ac02df by Tom Manshreck <shreck@google.com>:
Update Span description to remove "view" terminology, in light of recent clarification of the differences between a view and a span type.
PiperOrigin-RevId: 316734382
--
361b87d55b1809a5da3f72a996686f27b3d630c2 by Evan Brown <ezb@google.com>:
Allow for explicit conversion of values in btree range constructor/insert.
PiperOrigin-RevId: 316725951
GitOrigin-RevId: 1f80d4f1cb8847545627a2d0b18f697c4a763292
Change-Id: I74e2b095bc24710b27ed63ed94a50ef8f0fc897f
--
f012012ef78234a6a4585321b67d7b7c92ebc266 by Laramie Leavitt <lar@google.com>:
Slight restructuring of absl/random/internal randen implementation.
Convert round-keys.inc into randen_round_keys.cc file.
Consistently use a 128-bit pointer type for internal method parameters. This allows simpler pointer arithmetic in C++ & permits removal of some constants and casts.
Remove some redundancy in comments & constexpr variables. Specifically, all references to Randen algorithm parameters use RandenTraits; duplication in RandenSlow removed.
PiperOrigin-RevId: 312190313
--
dc8b42e054046741e9ed65335bfdface997c6063 by Abseil Team <absl-team@google.com>:
Internal change.
PiperOrigin-RevId: 312167304
--
f13d248fafaf206492c1362c3574031aea3abaf7 by Matthew Brown <matthewbr@google.com>:
Cleanup StrFormat extensions a little.
PiperOrigin-RevId: 312166336
--
9d9117589667afe2332bb7ad42bc967ca7c54502 by Derek Mauro <dmauro@google.com>:
Internal change
PiperOrigin-RevId: 312105213
--
9a12b9b3aa0e59b8ee6cf9408ed0029045543a9b by Abseil Team <absl-team@google.com>:
Complete IGNORE_TYPE macro renaming.
PiperOrigin-RevId: 311999699
--
64756f20d61021d999bd0d4c15e9ad3857382f57 by Gennadiy Rozental <rogeeff@google.com>:
Switch to fixed bytes specific default value.
This fixes the Abseil Flags for big endian platforms.
PiperOrigin-RevId: 311844448
--
bdbe6b5b29791dbc3816ada1828458b3010ff1e9 by Laramie Leavitt <lar@google.com>:
Change many distribution tests to use pcg_engine as a deterministic source of entropy.
It's reasonable to test that the BitGen itself has good entropy, however when testing the cross product of all random distributions x all the architecture variations x all submitted changes results in a large number of tests. In order to account for these failures while still using good entropy requires that our allowed sigma need to account for all of these independent tests.
Our current sigma values are too restrictive, and we see a lot of failures, so we have to either relax the sigma values or convert some of the statistical tests to use deterministic values.
This changelist does the latter.
PiperOrigin-RevId: 311840096
GitOrigin-RevId: f012012ef78234a6a4585321b67d7b7c92ebc266
Change-Id: Ic84886f38ff30d7d72c126e9b63c9a61eb729a1a
Abseil Team
Greg Falcon