diff options
author | Andy Polyakov <appro@openssl.org> | 2005-12-03 11:59:25 +0000 |
---|---|---|
committer | Andy Polyakov <appro@openssl.org> | 2005-12-03 11:59:25 +0000 |
commit | c6709b89c43f2ba084d347e42e14f3f66f8846bd (patch) | |
tree | 3375063a1355e1aa8d13da5268d25944208b9a53 /crypto | |
parent | 061bebc0d8889c93cb9a067679a7d675d7522929 (diff) |
Couple other benchmark comparisons for wp-x86_64.pl.
Diffstat (limited to 'crypto')
-rw-r--r-- | crypto/whrlpool/asm/wp-x86_64.pl | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/crypto/whrlpool/asm/wp-x86_64.pl b/crypto/whrlpool/asm/wp-x86_64.pl index cbdbb8ea7e..3855382036 100644 --- a/crypto/whrlpool/asm/wp-x86_64.pl +++ b/crypto/whrlpool/asm/wp-x86_64.pl @@ -12,9 +12,10 @@ # to 32-bit MMX version executed on same CPU. So why did I bother? # Well, it's faster than gcc 3.3.2 generated code by over 50%, and # over 80% faster than PathScale 1.4, an "ambitious" commercial -# compiler. What is it with x86_64 compilers? It's not the first -# example when they fail to generate more optimal code, when I -# believe they had *all* chances to... +# compiler. Furthermore it surpasses gcc 3.4.3 by 170% and Sun Studio +# 10 - by 360%[!]... What is it with x86_64 compilers? It's not the +# first example when they fail to generate more optimal code, when +# I believe they had *all* chances to... # # Note that register and stack frame layout are virtually identical # to 32-bit MMX version, except that %r8-15 are used instead of |