aboutsummaryrefslogtreecommitdiff
path: root/absl/hash/internal/hash.cc
Commit message (Collapse)AuthorAgeFilesLines
* Optimize `absl::Hash` by making `LowLevelHash` faster.Shahriar Rouf2024-02-071-1/+1
| | | | | | | | | | | | | | | | Throughput of the 64 byte chunk loop inside `LowLevelHash` (or now in `LowLevelHashLenGt16`) gets limited by the loop carried dependency on `current_state`. By using 4 states instead of 2, we can reduce this duration by 1 cycle. On Skylake, it is reduced from 9 cycles to 8 cycles (12.5% faster asymptotically). To see the reduction in a simplified version of `LowLevelHash` implementation on Skylake: * Before: https://godbolt.org/z/Tcj9vsGax, llvm-mca (https://godbolt.org/z/3o78Msr63) shows 9 cycles / iteration. * After: https://godbolt.org/z/q4GM4EjPr, llvm-mca (https://godbolt.org/z/W5d1KEMzq) shows 8 cycles / iteration. * This CL is removing 1 xor (1 cycle) per iteration from the critical path. A block for 32 byte chunk is also added. Finally, just before returning, `Mix` is called 1 time instead of twice. PiperOrigin-RevId: 605090653 Change-Id: Ib7517ebb8bef7484066cd14cf41a943953e93377
* Export of internal Abseil changesAbseil Team2021-09-301-3/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -- 506fa3e10b3d8399ad937c32ecea26d1ad4e62bb by Abseil Team <absl-team@google.com>: Disable ABSL_HAVE_STD_IS_TRIVIALLY_ASSIGNABLE when GCC < 8.2.0 is used with libc++ PiperOrigin-RevId: 399707056 -- 656b7c7cee87f46a4bc7953618796f82da08e62c by Derek Mauro <dmauro@google.com>: Remove the MSVC flag implementation from flag.h to help clarify that methods on absl::Flag<T> are not part of the public API PiperOrigin-RevId: 399584678 -- a92a9bc156303bc663b84c4b704891ec8f67e333 by Abseil Team <absl-team@google.com>: Get rid of MemcpyIfAllowed while continuing to suppress erroneous warnings PiperOrigin-RevId: 399468864 -- 5f9a66895f707ba001fb51b88c0c6025f8c872a3 by Abseil Team <absl-team@google.com>: Use feature testing to check for availability of invoke_result. Feature testing should be available by C++20, which removes invoke_result. https://en.cppreference.com/w/cpp/feature_test PiperOrigin-RevId: 399447373 -- 946c0a502b4499dbfcabf1ab93ddde0048288fb4 by CJ Johnson <johnsoncj@google.com>: Add rvalue-reference qualifier to the Commit method on ConstructionTransaction PiperOrigin-RevId: 399442206 -- 726a4d036eff49aeb6fd0ca2b1775699b6844395 by Greg Falcon <gfalcon@google.com>: Internal change PiperOrigin-RevId: 399441870 -- 1df6d3f659b88dbac13c3d8e13db23bb3844ece2 by Abseil Team <absl-team@google.com>: Clang-format whitespace changes PiperOrigin-RevId: 399281271 -- 4a828cde95a07421d699ebac775b37810624214f by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 399234071 -- e520c72b34ba2f98668c889139001f8276243d31 by Greg Falcon <gfalcon@google.com>: Import of CCTZ from GitHub. PiperOrigin-RevId: 399233662 GitOrigin-RevId: 506fa3e10b3d8399ad937c32ecea26d1ad4e62bb Change-Id: I92b9176d2387c08eb167f9268efa78b55b8e09c2
* Export of internal Abseil changesAbseil Team2021-08-131-6/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -- 3a9b4e8e5ecba532db5cc4ac12d12660307ce9fb by Derek Mauro <dmauro@google.com>: Use the Bazel @platforms repository for platform constraints Fixes #1000 PiperOrigin-RevId: 390644226 -- b34e4d2f8a86b54bd483ec4c9c3dd781ad2d8b68 by Abseil Team <absl-team@google.com>: debugging: add some handling for RISC-V The RISC-V architecture uses a downward growing stack and can host Linux using ELF files. Adjust a few sites accordingly to indicate how to handle the RISC-V architecture. PiperOrigin-RevId: 390631894 -- 5fa3a0961bf3dd0799c048956a0128f7b8113f1e by Samuel Benzaquen <sbenza@google.com>: Rename the buffer hash function to LowLevelHash. Although it started as wyhash, it will depart from it so it does not make sense to keep the name. PiperOrigin-RevId: 390483506 -- 2e7867a2301d58ad4cd5abcaa5fd6f0db973ae7b by Abseil Team <absl-team@google.com>: This is an internal change. PiperOrigin-RevId: 390349746 GitOrigin-RevId: 3a9b4e8e5ecba532db5cc4ac12d12660307ce9fb Change-Id: I322c3762552a2107e6c6b108c25c01e5efa8aecd
* Export of internal Abseil changesAbseil Team2021-05-121-8/+6
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -- 9fc37c11b9e46287acef00ee06ed9adcba54dd13 by Greg Falcon <gfalcon@google.com>: Rename absl::hash_internal::HashState to absl::hash_internal::MixingHashState. Before this change, we had two classes named HashState: absl::HashState, the public API used for type erasure, and absl::hash_internal::HashState, the internal concrete implementation ordinarily used. The internal class used to be named `CityHashState`, but we renamed it to `HashState` it when we changed underlying hash implementation to wyhash. This inadvertent naming conflict made the code much harder to read, and this change intends to undo that. PiperOrigin-RevId: 373481959 -- 4aec55ffddebd085c239352a2e20721091f719a1 by Greg Falcon <gfalcon@google.com>: Introduce absl::HashOf(), a convenience wrapper around absl::Hash that calculates hashes from the values of its arguments. PiperOrigin-RevId: 373461406 -- 86b5fd8db50bbc8bd0aa9258523527381fe0445d by Abseil Team <absl-team@google.com>: Improve speed of BlockingCounter by making its most common path lock free. With the new implementation, the fast path of BlockingCounter::DecrementCount() is only a fetch_sub operation. This is most times much more efficient than the previous implementation (full mutex lock/unlock). As a matter of fact, in most actual usecases in practice, the waiter thread is already waiting on the Wait() call when DecrementCount() is called, which makes Mutex::Unlock() take the slow path as there's a waiter thread that it might need to wake up. PiperOrigin-RevId: 373394164 -- 65c876be5eac0cd32583ff8535ede4109d39cf3f by Martijn Vels <mvels@google.com>: Move the 'sample copied cord' logic into MaybeTrackCord(), This changes move the logic for selecting if a cord should remain being sampled from Cord to CordzInfo::MaybeTrackCord, and updates the documentation for the latter method. PiperOrigin-RevId: 373363168 -- e84410bd0aada293a81dfb82656c952e209e21fb by Martijn Vels <mvels@google.com>: Add check for the first call to cordz_should_profile() for each thread. This prevents the first cord of a newly created thread to be always sampled, which is a 'bad' kind of determinism for sampling. PiperOrigin-RevId: 373229768 -- bf09c589dc099ac8f4af780bf7e609c53c27574c by Samuel Benzaquen <sbenza@google.com>: Refactor the Flags structure into an enum. This gives us more control over the representation and allows for easier merging during parsing. PiperOrigin-RevId: 373163038 -- b947b0c51083b7b6508284b5d31819596c91729e by Derek Mauro <dmauro@google.com>: Fixes warnings about shadowed variables Fixes #956 PiperOrigin-RevId: 373158133 GitOrigin-RevId: 9fc37c11b9e46287acef00ee06ed9adcba54dd13 Change-Id: I91f35699f9bf439d1a870c6493946a310afe088c
* Export of internal Abseil changesAbseil Team2020-11-241-10/+25
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -- fcedaa5714efab8738446fa21620b827a40a3458 by Derek Mauro <dmauro@google.com>: Uses Wyhash in the implementation of absl::Hash for hashing sequences of more than 16 bytes on some platforms. Due to the per-process randomization of the seed used by absl::Hash, users should not notice this change, other than possibly getting better performance. This change only affects platforms where absl::uint128 is implemented with an intrinsic (and where sizeof(size_t)==8) since Wyhash relies on fast 128-bit multiplication for speed. PiperOrigin-RevId: 343956735 -- 085e108c760084f19caa21dbeb2118de2be3f8f0 by Abseil Team <absl-team@google.com>: Internal change for cord ring PiperOrigin-RevId: 343919274 -- 4c333278ad14d6692f203074b902506008ad624a by Jorg Brown <jorg@google.com>: Minimize strings_internal::StringConstant further, by removing the need for the compiler to instantiate a compile-time class function. PiperOrigin-RevId: 343878568 -- 71c3c8c7b7821b67997e3d5345aaec67f93f266f by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 343838259 GitOrigin-RevId: fcedaa5714efab8738446fa21620b827a40a3458 Change-Id: Ifb91895a82d11e743acd42fe97ab7fb70712b7df
* Export of internal Abseil changesAbseil Team2019-12-121-0/+2
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -- c99f979ad34f155fbeeea69b88bdc7458d89a21c by Derek Mauro <dmauro@google.com>: Remove a floating point division by zero test. This isn't testing behavior related to the library, and MSVC warns about it in opt mode. PiperOrigin-RevId: 285220804 -- 68b015491f0dbf1ab547994673281abd1f34cd4b by Gennadiy Rozental <rogeeff@google.com>: This CL introduces following changes to the class FlagImpl: * We eliminate the CommandLineFlagLocks struct. Instead callback guard and callback function are combined into a single CallbackData struct, while primary data lock is stored separately. * CallbackData member of class FlagImpl is initially set to be nullptr and is only allocated and initialized when a flag's callback is being set. For most flags we do not pay for the extra space and extra absl::Mutex now. * Primary data guard is stored in data_guard_ data member. This is a properly aligned character buffer of necessary size. During initialization of the flag we construct absl::Mutex in this space using placement new call. * We now avoid extra value copy after successful attempt to parse value out of string. Instead we swap flag's current value with tentative value we just produced. PiperOrigin-RevId: 285132636 -- ed45d118fb818969eb13094cf7827c885dfc562c by Tom Manshreck <shreck@google.com>: Change null-term* (and nul-term*) to NUL-term* in comments PiperOrigin-RevId: 285036610 -- 729619017944db895ce8d6d29c1995aa2e5628a5 by Derek Mauro <dmauro@google.com>: Use the Posix implementation of thread identity on MinGW. Some versions of MinGW suffer from thread_local bugs. PiperOrigin-RevId: 285022920 -- 39a25493503c76885bc3254c28f66a251c5b5bb0 by Greg Falcon <gfalcon@google.com>: Implementation detail change. Add further ABSL_NAMESPACE_BEGIN and _END annotation macros to files in Abseil. PiperOrigin-RevId: 285012012 GitOrigin-RevId: c99f979ad34f155fbeeea69b88bdc7458d89a21c Change-Id: I4c85d3704e45d11a9ac50d562f39640a6adbedc1
* Export of internal Abseil changesAbseil Team2019-11-051-0/+30
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | -- 074a799119ac881b8b8ce59ef7a3166d1aa025ac by Tom Manshreck <shreck@google.com>: nit: Add return info for StrCat PiperOrigin-RevId: 278647298 -- d58a2a39ab6f50266cc695506ba2e86bdb45d795 by Mark Barolak <mbar@google.com>: Stop suppressing no-nested-anon-types warnings because there aren't actually any warnings to suppress. PiperOrigin-RevId: 278440548 -- 445051bd280b9a6f608a8c80b3d7cafcc1377a03 by Abseil Team <absl-team@google.com>: ResetThreadIdentity does not need to clear identity->waiter_state. ResetThreadIdentity is only called by NewThreadIdentity. NewThreadIdentity is only called by CreateThreadIdentity. CreateThreadIdentity calls PerThreadSem::Init, which initializes identity->waiter_state, immediately after calling NewThreadIdentity. Therefore ResetThreadIdentity does not need to clear identity->waiter_state. PiperOrigin-RevId: 278429844 -- c2079b664d92be40d5e365abcca4e9b3505a75a6 by Abseil Team <absl-team@google.com>: Delete the f->header.magic check in LowLevelAlloc::Free(). The f->header.magic check in LowLevelAlloc::Free() is redundant, because AddToFreeList() will immediately perform the same check. Also fix a typo in the comment that documents the lock requirements for Next(). The comment should say "L >= arena->mu", which is equivalent to EXCLUSIVE_LOCKS_REQUIRED(arena->mu). NOTE: LowLevelAlloc::Free() performs the f->header.magic check without holding the arena lock. This may have caused the TSAN data race warning reported in bug 143697235. PiperOrigin-RevId: 278414140 -- 5534f35ce677165700117d868f51607ed1f0d73b by Greg Falcon <gfalcon@google.com>: Add an internal (unsupported) PiecewiseCombiner class to allow hashing buffers piecewise. PiperOrigin-RevId: 278388902 GitOrigin-RevId: 074a799119ac881b8b8ce59ef7a3166d1aa025ac Change-Id: I61734850cbbb01c7585e8c736a5bb56e416512a8
* Changed HTTP URLs to HTTPS where possible (#270)nik72732019-03-081-1/+1
|
* Export of internal Abseil changes.Abseil Team2018-09-271-0/+23
-- 4eacae3ff1b14b1d309e8092185bc10e8a6203cf by Derek Mauro <dmauro@google.com>: Release SwissTable - a fast, efficient, cache-friendly hash table. https://www.youtube.com/watch?v=ncHmEUmJZf4 PiperOrigin-RevId: 214816527 -- df8c3dfab3cfb2f4365909a84d0683b193cfbb11 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 214785288 -- 1eabd5266bbcebc33eecc91e5309b751856a75c8 by Abseil Team <absl-team@google.com>: Internal change PiperOrigin-RevId: 214722931 -- 2ebbfac950f83146b46253038e7dd7dcde9f2951 by Derek Mauro <dmauro@google.com>: Internal change PiperOrigin-RevId: 214701684 GitOrigin-RevId: 4eacae3ff1b14b1d309e8092185bc10e8a6203cf Change-Id: I9ba64e395b22ad7863213d157b8019b082adc19d
generated by cgit v1.2.3 (git 2.39.1) at 2025-04-14 16:51:10 +0000