[PATCH v4 0/6] Introduce x86 assembly accelerated implementation for SM3 algorithm

Tianjia Zhang posted 6 patches 4 years, 5 months ago
arch/arm64/crypto/Kconfig        |   2 +-
arch/arm64/crypto/sm3-ce-glue.c  |  28 +-
arch/x86/crypto/Makefile         |   3 +
arch/x86/crypto/sm3-avx-asm_64.S | 517 +++++++++++++++++++++++++++++++
arch/x86/crypto/sm3_avx_glue.c   | 134 ++++++++
crypto/Kconfig                   |  16 +-
crypto/sm2.c                     |  38 +--
crypto/sm3_generic.c             | 142 +--------
crypto/tcrypt.c                  |  14 +-
include/crypto/sm3.h             |  34 +-
lib/crypto/Kconfig               |   3 +
lib/crypto/Makefile              |   3 +
lib/crypto/sm3.c                 | 246 +++++++++++++++
13 files changed, 1013 insertions(+), 167 deletions(-)
create mode 100644 arch/x86/crypto/sm3-avx-asm_64.S
create mode 100644 arch/x86/crypto/sm3_avx_glue.c
create mode 100644 lib/crypto/sm3.c
[PATCH v4 0/6] Introduce x86 assembly accelerated implementation for SM3 algorithm
Posted by Tianjia Zhang 4 years, 5 months ago
This series of patches creates an stand-alone library for SM3 hash
algorithm in the lib/crypto directory, and makes the implementations
that originally depended on sm3-generic depend on the stand-alone SM3
library, which also includes sm3-generic itself.

On this basis, the AVX assembly acceleration implementation of SM3
algorithm is introduced, the main algorithm implementation based on
SM3 AES/BMI2 accelerated work by libgcrypt at:
https://gnupg.org/software/libgcrypt/index.html

From the performance benchmark data, the performance improvement of
SM3 algorithm after AVX optimization can reach up to 38%.

---
v4 changes:
 - Rebase on latest cryptodev-2.6/master
 - Fix the compilation error of arm64/sm3

v3 changes:
 - update git commit message for patch 01

v2 changes:
 - x86/sm3: Change K macros to signed decimal and use LEA and 32-bit offset

Tianjia Zhang (6):
  crypto: sm3 - create SM3 stand-alone library
  crypto: arm64/sm3-ce - make dependent on sm3 library
  crypto: sm2 - make dependent on sm3 library
  crypto: sm3 - make dependent on sm3 library
  crypto: x86/sm3 - add AVX assembly implementation
  crypto: tcrypt - add asynchronous speed test for SM3

 arch/arm64/crypto/Kconfig        |   2 +-
 arch/arm64/crypto/sm3-ce-glue.c  |  28 +-
 arch/x86/crypto/Makefile         |   3 +
 arch/x86/crypto/sm3-avx-asm_64.S | 517 +++++++++++++++++++++++++++++++
 arch/x86/crypto/sm3_avx_glue.c   | 134 ++++++++
 crypto/Kconfig                   |  16 +-
 crypto/sm2.c                     |  38 +--
 crypto/sm3_generic.c             | 142 +--------
 crypto/tcrypt.c                  |  14 +-
 include/crypto/sm3.h             |  34 +-
 lib/crypto/Kconfig               |   3 +
 lib/crypto/Makefile              |   3 +
 lib/crypto/sm3.c                 | 246 +++++++++++++++
 13 files changed, 1013 insertions(+), 167 deletions(-)
 create mode 100644 arch/x86/crypto/sm3-avx-asm_64.S
 create mode 100644 arch/x86/crypto/sm3_avx_glue.c
 create mode 100644 lib/crypto/sm3.c

-- 
2.34.1

Re: [PATCH v4 0/6] Introduce x86 assembly accelerated implementation for SM3 algorithm
Posted by Herbert Xu 4 years, 5 months ago
On Fri, Jan 07, 2022 at 08:06:54PM +0800, Tianjia Zhang wrote:
> This series of patches creates an stand-alone library for SM3 hash
> algorithm in the lib/crypto directory, and makes the implementations
> that originally depended on sm3-generic depend on the stand-alone SM3
> library, which also includes sm3-generic itself.
> 
> On this basis, the AVX assembly acceleration implementation of SM3
> algorithm is introduced, the main algorithm implementation based on
> SM3 AES/BMI2 accelerated work by libgcrypt at:
> https://gnupg.org/software/libgcrypt/index.html
> 
> >From the performance benchmark data, the performance improvement of
> SM3 algorithm after AVX optimization can reach up to 38%.
> 
> ---
> v4 changes:
>  - Rebase on latest cryptodev-2.6/master
>  - Fix the compilation error of arm64/sm3
> 
> v3 changes:
>  - update git commit message for patch 01
> 
> v2 changes:
>  - x86/sm3: Change K macros to signed decimal and use LEA and 32-bit offset
> 
> Tianjia Zhang (6):
>   crypto: sm3 - create SM3 stand-alone library
>   crypto: arm64/sm3-ce - make dependent on sm3 library
>   crypto: sm2 - make dependent on sm3 library
>   crypto: sm3 - make dependent on sm3 library
>   crypto: x86/sm3 - add AVX assembly implementation
>   crypto: tcrypt - add asynchronous speed test for SM3
> 
>  arch/arm64/crypto/Kconfig        |   2 +-
>  arch/arm64/crypto/sm3-ce-glue.c  |  28 +-
>  arch/x86/crypto/Makefile         |   3 +
>  arch/x86/crypto/sm3-avx-asm_64.S | 517 +++++++++++++++++++++++++++++++
>  arch/x86/crypto/sm3_avx_glue.c   | 134 ++++++++
>  crypto/Kconfig                   |  16 +-
>  crypto/sm2.c                     |  38 +--
>  crypto/sm3_generic.c             | 142 +--------
>  crypto/tcrypt.c                  |  14 +-
>  include/crypto/sm3.h             |  34 +-
>  lib/crypto/Kconfig               |   3 +
>  lib/crypto/Makefile              |   3 +
>  lib/crypto/sm3.c                 | 246 +++++++++++++++
>  13 files changed, 1013 insertions(+), 167 deletions(-)
>  create mode 100644 arch/x86/crypto/sm3-avx-asm_64.S
>  create mode 100644 arch/x86/crypto/sm3_avx_glue.c
>  create mode 100644 lib/crypto/sm3.c

All applied.  Thanks.
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt