Messages in this thread | | | Date | Mon, 20 Dec 2021 20:03:47 +0200 | From | Jussi Kivilinna <> | Subject | Re: [PATCH 5/6] crypto: x86/sm3 - add AVX assembly implementation |
| |
On 20.12.2021 10.22, Tianjia Zhang wrote: > This patch adds AVX assembly accelerated implementation of SM3 secure > hash algorithm. From the benchmark data, compared to pure software > implementation sm3-generic, the performance increase is up to 38%. > > The main algorithm implementation based on SM3 AES/BMI2 accelerated > work by libgcrypt at: > https://gnupg.org/software/libgcrypt/index.html > > Benchmark on Intel i5-6200U 2.30GHz, performance data of two > implementations, pure software sm3-generic and sm3-avx acceleration. > The data comes from the 326 mode and 422 mode of tcrypt. The abscissas > are different lengths of per update. The data is tabulated and the > unit is Mb/s: > > update-size | 16 64 256 1024 2048 4096 8192 > -------------------------------------------------------------------- > sm3-generic | 105.97 129.60 182.12 189.62 188.06 193.66 194.88 > sm3-avx | 119.87 163.05 244.44 260.92 257.60 264.87 265.88 > > Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com> > --- > arch/x86/crypto/Makefile | 3 + > arch/x86/crypto/sm3-avx-asm_64.S | 521 +++++++++++++++++++++++++++++++ > arch/x86/crypto/sm3_avx_glue.c | 134 ++++++++ > crypto/Kconfig | 13 + > 4 files changed, 671 insertions(+) > create mode 100644 arch/x86/crypto/sm3-avx-asm_64.S > create mode 100644 arch/x86/crypto/sm3_avx_glue.c > > diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile > index f307c93fc90a..7cbe860f6201 100644 > --- a/arch/x86/crypto/Makefile > +++ b/arch/x86/crypto/Makefile > @@ -88,6 +88,9 @@ nhpoly1305-avx2-y := nh-avx2-x86_64.o nhpoly1305-avx2-glue.o > > obj-$(CONFIG_CRYPTO_CURVE25519_X86) += curve25519-x86_64.o > > +obj-$(CONFIG_CRYPTO_SM3_AVX_X86_64) += sm3-avx-x86_64.o > +sm3-avx-x86_64-y := sm3-avx-asm_64.o sm3_avx_glue.o > + > obj-$(CONFIG_CRYPTO_SM4_AESNI_AVX_X86_64) += sm4-aesni-avx-x86_64.o > sm4-aesni-avx-x86_64-y := sm4-aesni-avx-asm_64.o sm4_aesni_avx_glue.o > > diff --git a/arch/x86/crypto/sm3-avx-asm_64.S b/arch/x86/crypto/sm3-avx-asm_64.S > new file mode 100644 > index 000000000000..e7a9a37f3609 > --- /dev/null > +++ b/arch/x86/crypto/sm3-avx-asm_64.S > @@ -0,0 +1,521 @@ > +/* SPDX-License-Identifier: GPL-2.0-or-later */ > +/* > + * SM3 AVX accelerated transform. > + * specified in: https://datatracker.ietf.org/doc/html/draft-sca-cfrg-sm3-02 > + * > + * Copyright (C) 2021 Jussi Kivilinna <jussi.kivilinna@iki.fi> > + * Copyright (C) 2021 Tianjia Zhang <tianjia.zhang@linux.alibaba.com> > + */ <snip> > + > +#define R(i, a, b, c, d, e, f, g, h, round, widx, wtype) \ > + /* rol(a, 12) => t0 */ \ > + roll3mov(12, a, t0); /* rorxl here would reduce perf by 6% on zen3 */ \ > + /* rol (t0 + e + t), 7) => t1 */ \ > + addl3(t0, e, t1); \ > + addl $K##round, t1; \
It's better to use "leal K##round(t0, e, 1), t1;" here and fix K0-K63 macros instead as I noted at libgcrypt mailing-list: https://lists.gnupg.org/pipermail/gcrypt-devel/2021-December/005209.html
-Jussi
| |