crypto: serpent - add 8-way parallel x86_64/SSE2 assembler implementation
authorJussi Kivilinna <jussi.kivilinna@mbnet.fi>
Wed, 9 Nov 2011 14:26:25 +0000 (16:26 +0200)
committerHerbert Xu <herbert@gondor.apana.org.au>
Mon, 21 Nov 2011 08:13:23 +0000 (16:13 +0800)
commit937c30d7f560210b0163035edd42b2aef78fed9e
treec47348474ca6cdda0a87c95f3a6831c732f27b4d
parentd19978f58745e586d9385d306d557e7c785abe23
crypto: serpent - add 8-way parallel x86_64/SSE2 assembler implementation

Patch adds x86_64/SSE2 assembler implementation of serpent cipher. Assembler
functions crypt data in eigth block chunks (two 4 block chunk SSE2 operations
in parallel to improve performance on out-of-order CPUs). Glue code is based
on one from AES-NI implementation, so requests from irq context are redirected
to cryptd.

v2:
 - add missing include of linux/module.h
   (appearently crypto.h used to include module.h, which changed for 3.2 by
    commit 7c926402a7e8c9b279968fd94efec8700ba3859e)

Patch has been tested with tcrypt and automated filesystem tests.

Tcrypt benchmarks results (serpent-sse2/serpent_generic speed ratios):

AMD Phenom II 1055T (fam:16, model:10):

size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec
16B     1.03x   1.01x   1.03x   1.05x   1.00x   0.99x
64B     1.00x   1.01x   1.02x   1.04x   1.02x   1.01x
256B    2.34x   2.41x   0.99x   2.43x   2.39x   2.40x
1024B   2.51x   2.57x   1.00x   2.59x   2.56x   2.56x
8192B   2.50x   2.54x   1.00x   2.55x   2.57x   2.57x

Intel Celeron T1600 (fam:6, model:15, step:13):

size    ecb-enc ecb-dec cbc-enc cbc-dec ctr-enc ctr-dec
16B     0.97x   0.97x   1.01x   1.01x   1.01x   1.02x
64B     1.00x   1.00x   1.00x   1.02x   1.01x   1.01x
256B    3.41x   3.35x   1.00x   3.39x   3.42x   3.44x
1024B   3.75x   3.72x   0.99x   3.74x   3.75x   3.75x
8192B   3.70x   3.68x   0.99x   3.68x   3.69x   3.69x

Full output:
 http://koti.mbnet.fi/axh/kernel/crypto/phenom-ii-1055t/serpent-generic.txt
 http://koti.mbnet.fi/axh/kernel/crypto/phenom-ii-1055t/serpent-sse2.txt
 http://koti.mbnet.fi/axh/kernel/crypto/celeron-t1600/serpent-generic.txt
 http://koti.mbnet.fi/axh/kernel/crypto/celeron-t1600/serpent-sse2.txt

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
arch/x86/crypto/Makefile
arch/x86/crypto/serpent-sse2-x86_64-asm_64.S [new file with mode: 0644]
arch/x86/crypto/serpent_sse2_glue.c [new file with mode: 0644]
arch/x86/include/asm/serpent.h [new file with mode: 0644]
crypto/Kconfig
crypto/testmgr.c