[0/2] crypto: arm64/aes-ce - implement 5-way interleave for some modes

Message

Ard Biesheuvel June 24, 2019, 5:38 p.m. UTC

As it turns out, even a 4-way interleave is not sufficient to saturate
the ThunderX2 pipeline with AES instructions, so this series implements
5-way interleave for modes that can be modified without running out of
registers to maintain chaining mode state across the encryption operation,
i.e., ECB, CBC-decryption and CTR.

Ard Biesheuvel (2):
  crypto: arm64/aes-ce - add 5 way interleave routines
  crypto: arm64/aes-ce - implement 5 way interleave for ECB, CBC and CTR

 arch/arm64/crypto/aes-ce.S    |  60 ++++++----
 arch/arm64/crypto/aes-modes.S | 118 +++++++++++++++-----
 arch/arm64/crypto/aes-neon.S  |  48 +-------
 3 files changed, 127 insertions(+), 99 deletions(-)

Comments

Herbert Xu July 3, 2019, 2:27 p.m. UTC | #1

On Mon, Jun 24, 2019 at 07:38:29PM +0200, Ard Biesheuvel wrote:
> As it turns out, even a 4-way interleave is not sufficient to saturate
> the ThunderX2 pipeline with AES instructions, so this series implements
> 5-way interleave for modes that can be modified without running out of
> registers to maintain chaining mode state across the encryption operation,
> i.e., ECB, CBC-decryption and CTR.
> 
> Ard Biesheuvel (2):
>   crypto: arm64/aes-ce - add 5 way interleave routines
>   crypto: arm64/aes-ce - implement 5 way interleave for ECB, CBC and CTR
> 
>  arch/arm64/crypto/aes-ce.S    |  60 ++++++----
>  arch/arm64/crypto/aes-modes.S | 118 +++++++++++++++-----
>  arch/arm64/crypto/aes-neon.S  |  48 +-------
>  3 files changed, 127 insertions(+), 99 deletions(-)

All applied.  Thanks.