From 85e28dfa6ffe1133b9f6e170e54ac39e9580731b Mon Sep 17 00:00:00 2001 From: Andy Polyakov Date: Mon, 26 Jul 2010 21:54:21 +0000 Subject: [PATCH] ghash-ia64.pl: excuse myself from implementing "528B" variant. --- crypto/modes/asm/ghash-ia64.pl | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/crypto/modes/asm/ghash-ia64.pl b/crypto/modes/asm/ghash-ia64.pl index d24a44fea2..299848ff85 100755 --- a/crypto/modes/asm/ghash-ia64.pl +++ b/crypto/modes/asm/ghash-ia64.pl @@ -20,6 +20,15 @@ # something else sha1-ia64.pl module processes one byte in 6.0 cycles. # On Itanium GHASH should run at ~8.5 cycles per byte. +# Note about "528B" variant. In Itanium 2 case it makes lesser sense +# to implement it for following reason. Because number of functional +# units is naturally limited, it's impossible to implement "528B" loop +# in 4 cycles, only in 5. This means that theoretically performance +# improvement can't be more than 20%, ~15% is more realistic. This +# is considered below justification level for implementing new code. +# Not to mention that on original Itanium it would actually run +# slower, spending >9 cycles per byte. + $output=shift and (open STDOUT,">$output" or die "can't open $output: $!"); if ($^O eq "hpux") { -- 2.25.1