diff options
author | Jussi Kivilinna <jussi.kivilinna@iki.fi> | 2013-10-26 15:00:48 +0300 |
---|---|---|
committer | Jussi Kivilinna <jussi.kivilinna@iki.fi> | 2013-10-28 11:29:36 +0200 |
commit | e214e8392671dd30e9c33260717b5e756debf3bf (patch) | |
tree | d8fc987d848a9d42012f81065ffe6fc486d43c65 /tests/Makefile.am | |
parent | ebc8abfcb09d6106fcfce40f240a513e276f46e9 (diff) | |
download | libgcrypt-e214e8392671dd30e9c33260717b5e756debf3bf.tar.gz |
Add new benchmarking utility, bench-slope
* tests/Makefile.am (TESTS): Add 'bench-slope'.
* tests/bench-slope.c: New.
--
Bench-slope is new benchmarking tool for libgcrypt for obtaining overheadless
cycles/byte speed of cipher and hash algorithms. Tool measures the time each
operation (hash/encrypt/decrypt/authentication) takes for different buffer
sizes of from ~0kB to ~4kB and calculates the slope for these data points.
The default output is then given as nanosecs/byte and mebibytes/sec. If user
provides the speed of used CPU, tool also outputs cycles/byte result (CPU-Ghz *
ns/B = c/B).
Output without CPU speed (with ARM Cortex-A8):
$ tests/bench-slope hash
Hash:
| nanosecs/byte mebibytes/sec cycles/byte
MD5 | 7.35 ns/B 129.7 MiB/s - c/B
SHA1 | 12.30 ns/B 77.53 MiB/s - c/B
RIPEMD160 | 15.96 ns/B 59.77 MiB/s - c/B
TIGER192 | 55.55 ns/B 17.17 MiB/s - c/B
SHA256 | 24.38 ns/B 39.12 MiB/s - c/B
SHA384 | 34.24 ns/B 27.86 MiB/s - c/B
SHA512 | 34.19 ns/B 27.90 MiB/s - c/B
SHA224 | 24.38 ns/B 39.12 MiB/s - c/B
MD4 | 5.68 ns/B 168.0 MiB/s - c/B
CRC32 | 9.26 ns/B 103.0 MiB/s - c/B
CRC32RFC1510 | 9.20 ns/B 103.6 MiB/s - c/B
CRC24RFC2440 | 87.31 ns/B 10.92 MiB/s - c/B
WHIRLPOOL | 253.3 ns/B 3.77 MiB/s - c/B
TIGER | 55.55 ns/B 17.17 MiB/s - c/B
TIGER2 | 55.55 ns/B 17.17 MiB/s - c/B
GOSTR3411_94 | 212.0 ns/B 4.50 MiB/s - c/B
STRIBOG256 | 630.1 ns/B 1.51 MiB/s - c/B
STRIBOG512 | 630.1 ns/B 1.51 MiB/s - c/B
=
With CPU speed (with Intel i5-4570, 3.2Ghz when turbo-boost disabled):
$ tests/bench-slope --cpu-mhz 3201 cipher arcfour blowfish aes
Cipher:
ARCFOUR | nanosecs/byte mebibytes/sec cycles/byte
STREAM enc | 2.43 ns/B 392.1 MiB/s 7.79 c/B
STREAM dec | 2.44 ns/B 390.2 MiB/s 7.82 c/B
=
BLOWFISH | nanosecs/byte mebibytes/sec cycles/byte
ECB enc | 7.62 ns/B 125.2 MiB/s 24.38 c/B
ECB dec | 7.63 ns/B 125.0 MiB/s 24.43 c/B
CBC enc | 9.18 ns/B 103.9 MiB/s 29.38 c/B
CBC dec | 2.60 ns/B 366.2 MiB/s 8.34 c/B
CFB enc | 9.17 ns/B 104.0 MiB/s 29.35 c/B
CFB dec | 2.66 ns/B 358.1 MiB/s 8.53 c/B
OFB enc | 8.97 ns/B 106.3 MiB/s 28.72 c/B
OFB dec | 8.97 ns/B 106.3 MiB/s 28.71 c/B
CTR enc | 2.60 ns/B 366.5 MiB/s 8.33 c/B
CTR dec | 2.60 ns/B 367.1 MiB/s 8.32 c/B
=
AES | nanosecs/byte mebibytes/sec cycles/byte
ECB enc | 0.439 ns/B 2173.0 MiB/s 1.40 c/B
ECB dec | 0.489 ns/B 1949.5 MiB/s 1.57 c/B
CBC enc | 1.64 ns/B 580.8 MiB/s 5.26 c/B
CBC dec | 0.219 ns/B 4357.6 MiB/s 0.701 c/B
CFB enc | 1.53 ns/B 623.6 MiB/s 4.90 c/B
CFB dec | 0.219 ns/B 4350.5 MiB/s 0.702 c/B
OFB enc | 1.51 ns/B 629.9 MiB/s 4.85 c/B
OFB dec | 1.51 ns/B 629.9 MiB/s 4.85 c/B
CTR enc | 0.288 ns/B 3308.5 MiB/s 0.923 c/B
CTR dec | 0.288 ns/B 3316.9 MiB/s 0.920 c/B
CCM enc | 1.93 ns/B 493.8 MiB/s 6.18 c/B
CCM dec | 1.93 ns/B 494.0 MiB/s 6.18 c/B
CCM auth | 1.64 ns/B 580.1 MiB/s 5.26 c/B
=
Note: It's highly recommented to disable turbo-boost and dynamic CPU frequency
features when making these kind of measurements to reduce variance.
Note: The results are maximum performance for each operation; the actual speed
in application depends on various matters, such as: used buffer sizes, cache
usage, etc.
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
Diffstat (limited to 'tests/Makefile.am')
-rw-r--r-- | tests/Makefile.am | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/tests/Makefile.am b/tests/Makefile.am index ac84e751..c9ba5f45 100644 --- a/tests/Makefile.am +++ b/tests/Makefile.am @@ -24,8 +24,8 @@ TESTS = version mpitests tsexp t-convert \ fips186-dsa aeswrap pkcs1v2 random dsa-rfc6979 t-ed25519 -# The last test to run. -TESTS += benchmark +# The last tests to run. +TESTS += benchmark bench-slope # Need to include ../src in addition to top_srcdir because gcrypt.h is |