Huge Speedups For CRC32C With Modern AVX-512 CPUs Merged To Linux 6.17
The CRC32C cyclic redundancy check code path within the Linux kernel for error detection is much, much faster with the in-development Linux 6.17 kernel when running on modern Intel and AMD AVX-512 processors.
AVX-512 processors with the VPCLMULQDQ instruction will benefit from faster CRC32C performance on long messages with the upcoming Linux 6.17 release. The patches previously covered on Phoronix earlier this month within Linux Kernel Patches Speed-Up CRC32 Performance For CPUs With "Good" AVX-512 have managed to make it into Linux 6.17 Git as of Monday.
The patch with the VPCLMULQDQ optimization for CRC32C notes AMD Zen 4 / Zen 5 being in better shape with their AVX-512 implementation as well as there being room for further improvements.
The shiny improvement was merged as part of the CRC updates for Linux 6.17.
Eric Biggers of Google had also sent out another pull request for the crypto library updates for Linux 6.17 that include converting FSVERITY and AppArmor to using the SHA-2 library functions rather than the crypto_shash API. That change should yield slightly faster performance and being simpler code-wise.
AVX-512 processors with the VPCLMULQDQ instruction will benefit from faster CRC32C performance on long messages with the upcoming Linux 6.17 release. The patches previously covered on Phoronix earlier this month within Linux Kernel Patches Speed-Up CRC32 Performance For CPUs With "Good" AVX-512 have managed to make it into Linux 6.17 Git as of Monday.
👁 CRC32C Linux 6.17 benchmarks
The patch with the VPCLMULQDQ optimization for CRC32C notes AMD Zen 4 / Zen 5 being in better shape with their AVX-512 implementation as well as there being room for further improvements.
The shiny improvement was merged as part of the CRC updates for Linux 6.17.
Eric Biggers of Google had also sent out another pull request for the crypto library updates for Linux 6.17 that include converting FSVERITY and AppArmor to using the SHA-2 library functions rather than the crypto_shash API. That change should yield slightly faster performance and being simpler code-wise.