Re: More performance tests of different AES implementations

From: brian gladman (brg_at_gladman.plus.com)
Date: 07/06/03 

Date: Sun, 6 Jul 2003 15:51:22 +0100

"Mok-Kong Shen" <mok-kong.shen@t-online.de> wrote in message
news:3F08219E.829B6169@t-online.de...

[snip]
> Since you have much experience and have already done
> extensive work in timing measurements, it would be very
> nice, if you could do some experiments to comare your
> above approaches of yours with the one simply using
> the function clock().

Here is a run that includes the timing of my encryption code at 23.9 cycles
per byte:

Version EKS(cycles/key) ENC(cycles/byte) DKS(cycles/key) DEC(cycles/byte)

<empty> 40( 0.0%) 2.5( 0.0%) 40( 0.0%) 2.5( 0.0%)
devine 926( 0.8%) 29.2( 0.6%) 926( 0.4%) 30.6( 1.1%)
gladman 195( 3.0%) 23.9( 0.7%) 327( 1.9%) 26.4( 3.5%)
gnupg 1259( 1.3%) 41.3( 0.7%) 444( 0.4%) 50.8( 0.7%)
mks 1577( 0.3%) 35.0( 0.8%) 1891( 0.5%) 34.6( 1.9%)
openssl 337( 0.3%) 27.4( 1.0%) 868( 0.4%) 25.9( 0.7%)
tom 784( 1.0%) 27.7( 0.2%) 786( 0.8%) 32.9( 0.7%)

Here is the cycles per byte measured with clock() for various numbers of
calls to encrypt():

    100000: 22.50 1000000: 24.75 10000000: 24.19

As can be seen this is converging towards 23.9 cycles per byte.

    Brian Gladman