diff options
| author |  Connal de Souza <connaldesouza@google.com> | 2023-09-26 14:13:18 -0700 |
|---|---|---|
| committer |  Copybara-Service <copybara-worker@google.com> | 2023-09-26 14:13:55 -0700 |
| commit | ac364eb9d07ff4d8c7fb8d848b66398559220d04 (patch) | |
| tree | 1e001cba5a64ac395b77fa561333487729683baa | |
| parent | 2fa24cc42f9103f2af71b3df858e8b6814ff1fd4 (diff) | |
Optimize CRC32 for Ampere Siryn
Siryn's crc32 instruction seems to have latency 3 and throughput 1, which makes the optimal ratio of pmull and crc streams close to that of tested x86 machines. Up to +120% faster for large inputs.
PiperOrigin-RevId: 568645559
Change-Id: I86b85b1b2a5d4fb3680c516c4c9044238b20fe61
| -rw-r--r-- | absl/crc/internal/crc_x86_arm_combined.cc | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/absl/crc/internal/crc_x86_arm_combined.cc b/absl/crc/internal/crc_x86_arm_combined.cc index 4847bc83..dfb3b325 100644 --- a/absl/crc/internal/crc_x86_arm_combined.cc +++ b/absl/crc/internal/crc_x86_arm_combined.cc @@ -636,6 +636,9 @@ CRCImpl* TryNewCRC32AcceleratedX86ARMCombined() { case CpuType::kArmNeoverseN1: return new CRC32AcceleratedX86ARMCombinedMultipleStreams< 1, 1, CutoffStrategy::Unroll64CRC>(); + case CpuType::kAmpereSiryn: + return new CRC32AcceleratedX86ARMCombinedMultipleStreams< + 3, 2, CutoffStrategy::Fold3>(); #if defined(__aarch64__) default: // Not all ARM processors support the needed instructions, so check here |