GitLab

Daniel P. Berrange authored Aug 03, 2017 and Niels Möller committed Aug 26, 2017

The nettle-benchmark program currently uses the openssl low level
cipher APIs for benchmarking. This means it always runs the generic
software implementation, never able to take advantage of impls
optimized for new hardware (eg AES-NI).

Rewriting it to use the higher EVP APIs means we can use the same
code for all ciphers, and automatically trigger hardware optimized
versions, giving a fairer comparison against openssl as commonly
used in applications.

Use of the generic openssl impl can still be forced by setting an
env variable  OPENSSL_ia32cap="~0x200000200000000"
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>

Do benchmarking of arcfour, salsa20 and chacha via time_aead.

Martin Storsjö authored Mar 25, 2013 and Niels Möller committed Mar 25, 2013

This avoids unconditionally using clock_gettime in
{ecc,hogweed}-benchmark, falling back to clock() in these
as well.

This makes building succeed on e.g. OS X.

Rev: nettle/examples/nettle-benchmark.c:1.22

(time_cipher): Also benchmark CTR mode.

Rev: nettle/examples/nettle-benchmark.c:1.21

renaming. Benchmark both encryption and hashing.
(time_gmac): ...old name.

Rev: nettle/examples/nettle-benchmark.c:1.20

(cgt_time_start): Likewise.
(cgt_time_end): Likewise.
(clock_time_start): Likewise.
(clock_time_end): Likewise.
(time_function): Read clock via function pointers time_start and
time_end, so we can select method at runtime.
(xalloc): Use die function.
(main): Choose timing function. If available, try clock_gettime,
and fall back to clock if it doesn't exist.

Rev: nettle/examples/nettle-benchmark.c:1.19

(TIME_END, TIME_START): Check return value from clock_gettime.

Rev: nettle/examples/nettle-benchmark.c:1.18

Rev: nettle/examples/nettle-benchmark.c:1.17

(main): Call time_gmac.

Rev: nettle/examples/nettle-benchmark.c:1.16