aria: add x86_64 GFNI/AVX512 accelerated implementation
* cipher/Makefile.am: Add 'aria-gfni-avx512-amd64.S'. * cipher/aria-gfni-avx512-amd64.S: New. * cipher/aria.c (USE_GFNI_AVX512): New. [USE_GFNI_AVX512] (MAX_PARALLEL_BLKS): New. (ARIA_context): Add 'use_gfni_avx512'. (_gcry_aria_gfni_avx512_ecb_crypt_blk64) (_gcry_aria_gfni_avx512_ctr_crypt_blk64) (aria_gfni_avx512_ecb_crypt_blk64) (aria_gfni_avx512_ctr_crypt_blk64): New. (aria_crypt_blocks) [USE_GFNI_AVX512]: Add 64 parallel block AVX512/GFNI processing. (_gcry_aria_ctr_enc) [USE_GFNI_AVX512]: Add 64 parallel block AVX512/GFNI processing. (aria_setkey): Enable GFNI/AVX512 based on HW features. * configure.ac: Add 'aria-gfni-avx512-amd64.lo'.
This patch adds AVX512/GFNI accelerated ARIA block cipher
implementation for libgcrypt. This implementation is based on
work by Taehee Yoo, with following notable changes:
- Integration to libgcrypt, use of 'aes-common-amd64.h'.
- Use round loop instead of unrolling for smaller code size and increased performance.
- Use stack for temporary storage instead of external buffers.
- Add byte-addition fast path for CTR.
Benchmark on AMD Ryzen 9 7900X (zen4, turbo-freq off):
GFNI/AVX512:
ARIA128 | nanosecs/byte mebibytes/sec cycles/byte auto Mhz
ECB enc | 0.203 ns/B 4703 MiB/s 0.953 c/B 4700 ECB dec | 0.204 ns/B 4675 MiB/s 0.959 c/B 4700 CTR enc | 0.207 ns/B 4609 MiB/s 0.973 c/B 4700 CTR dec | 0.207 ns/B 4608 MiB/s 0.973 c/B 4700
Benchmark on Intel Core i3-1115G4 (tiger-lake, turbo-freq off):
GFNI/AVX512:
ARIA128 | nanosecs/byte mebibytes/sec cycles/byte auto Mhz
ECB enc | 0.362 ns/B 2635 MiB/s 1.08 c/B 2992 ECB dec | 0.361 ns/B 2639 MiB/s 1.08 c/B 2992 CTR enc | 0.362 ns/B 2633 MiB/s 1.08 c/B 2992 CTR dec | 0.362 ns/B 2633 MiB/s 1.08 c/B 2992
[v2]:
- Add byte-addition fast path for CTR.
Cc: Taehee Yoo <ap420073@gmail.com>
- Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>