diff options
author | Andy Polyakov <appro@openssl.org> | 2011-06-27 09:46:16 +0000 |
---|---|---|
committer | Andy Polyakov <appro@openssl.org> | 2011-06-27 09:46:16 +0000 |
commit | 0772f3b4f6a47aaa53e6cc2afdb0ec880d693c89 (patch) | |
tree | ea419fd2f1f698460f166834bba1ced4f903c5b5 /crypto | |
parent | 0a9a692e4e3be7e2ef5061e42e9907cae09c4cd2 (diff) |
rc4-x86_64.pl: commentary update.
Diffstat (limited to 'crypto')
-rwxr-xr-x | crypto/rc4/asm/rc4-x86_64.pl | 9 |
1 files changed, 6 insertions, 3 deletions
diff --git a/crypto/rc4/asm/rc4-x86_64.pl b/crypto/rc4/asm/rc4-x86_64.pl index 9c2daf40aa..e18e8a0008 100755 --- a/crypto/rc4/asm/rc4-x86_64.pl +++ b/crypto/rc4/asm/rc4-x86_64.pl @@ -86,14 +86,17 @@ # is better) and improvement coefficients relative to previous # version of this module are: # -# Opteron 5.3/+0% +# Opteron 5.3/+0%(*) # P4 6.5 -# Core2 6.2/+15%(*) +# Core2 6.2/+15%(**) # Westmere 4.2/+60% # Sandy Bridge 4.2/+120% # Atom 9.3/+80% # -# (*) Note that Core2 result is ~15% lower than corresponding result +# (*) But corresponding loop has less instructions, which should have +# positive effect on upcoming Bulldozer, which has one less ALU. +# For reference, Intel code runs at 6.8 cpb rate on Opteron. +# (**) Note that Core2 result is ~15% lower than corresponding result # for 32-bit code, meaning that it's possible to improve it, # but more than likely at the cost of the others (see rc4-586.pl # to get the idea)... |