| Commit message (Collapse) | Author | Age |
|
|
|
|
| |
PiperOrigin-RevId: 494587777
Change-Id: I41504edca6fcf750d52602fa84a33bc7fe5fbb48
|
|\
| |
| |
| |
| | |
PiperOrigin-RevId: 493386604
Change-Id: I289cb38b4a3da5760ab7ef3976d402d165d7e10f
|
|/ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
coverage of the accelerated CRC implementation and some differences
bewteen the internal and external implementation.
This change adds CI coverage to the
linux_clang-latest_libstdcxx_bazel.sh script assuming this script
always runs on machines of at least the Intel Haswell generation.
Fixes include:
* Remove the use of the deprecated xor operator on crc32c_t
* Remove #pragma unroll_completely, which isn't known by GCC or Clang:
https://godbolt.org/z/97j4vbacs
* Fixes for -Wsign-compare, -Wsign-conversion and -Wshorten-64-to-32
PiperOrigin-RevId: 491965029
Change-Id: Ic5e1f3a20f69fcd35fe81ebef63443ad26bf7931
|
|
|
|
|
| |
PiperOrigin-RevId: 491722639
Change-Id: Iff13661095d10c82599ad30f7220700825a78c9e
|
|
|
|
|
|
|
|
|
|
| |
std::array has a special-case to allow this
https://en.cppreference.com/w/cpp/container/array
Fixes #1332
PiperOrigin-RevId: 491703960
Change-Id: Ib83a1f0865448314e463e8ebf39ae3b842f762ea
|
|
|
|
|
| |
PiperOrigin-RevId: 491681300
Change-Id: I4ecdd3bf359cda7592b6c392a2fbb61b8394f71b
|
|
|
|
|
|
|
|
|
| |
The motivation is to explicitly remove and document dangerous
operations like adding crc32c_t to a set, because equality is not
enough to guarantee uniqueness.
PiperOrigin-RevId: 491656425
Change-Id: I7b4dadc1a59ea9861e6ec7a929d64b5746467832
|
|
|
|
|
|
|
|
|
|
|
|
| |
According to https://stackoverflow.com/a/68939636 it is safe to use
__m128i instead.
https://learn.microsoft.com/en-us/cpp/intrinsics/x86-intrinsics-list?view=msvc-170 also uses this type instead
Fixes #1330
PiperOrigin-RevId: 491427300
Change-Id: I4a1d44ac4d5e7c1e1ee063ff397935df118254a1
|
|
|
|
|
|
|
| |
Fixes #1329
PiperOrigin-RevId: 491372279
Change-Id: I93c094b06ece9cb9bdb39fd4541353e0344a1a57
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently we take generic/default code-path on AMD due to misspelling.
Mostly helps with crc+memcpy:
name old speed new speed delta
BM_Memcpy/1 156MB/s ± 1% 156MB/s ± 1% ~ (p=0.563 n=18+18)
BM_Memcpy/100 6.38GB/s ± 1% 6.50GB/s ± 1% +1.89% (p=0.000 n=19+19)
BM_Memcpy/10000 14.6GB/s ± 1% 21.7GB/s ± 0% +49.01% (p=0.000 n=20+19)
BM_Memcpy/500000 13.5GB/s ± 1% 19.9GB/s ± 0% +47.35% (p=0.000 n=18+17)
PiperOrigin-RevId: 490572650
Change-Id: Id7901321a23262c0ab62a2d82fae86cf42acf16d
|
|
|
|
|
|
|
| |
Using /arch:AVX on MSVC now uses the accelerated implementation
PiperOrigin-RevId: 490550573
Change-Id: I924259845f38ee41d15f23f95ad085ad664642b5
|
|
|
|
|
| |
PiperOrigin-RevId: 488373221
Change-Id: I1e30820188cc860ce4df8fddafa04de343ec46af
|
|
|
|
|
|
|
|
|
|
|
|
| |
This fixes the build on arm64 macOS.
Note that hardware acceleration is not yet enabled on arm64 when not
running under Linux.
Addresses the report from https://github.com/abseil/abseil-cpp/commit/1687dbf814eceb93de2d93f91b31acaab404091c#commitcomment-89529264
PiperOrigin-RevId: 487655295
Change-Id: I168dfc863c960d0b694b26dfcb85ff0fd0e95a1e
|
|
This implementation can advantage of hardware acceleration available
on common CPUs when using GCC and Clang. A future update may enable
this on MSVC as well.
PiperOrigin-RevId: 487327024
Change-Id: I99a8f1bcbdf25297e776537e23bd0a902e0818a1
|