ec/asm/ecp_nistz256-x86_64.pl: update commentary with before-after performance data.
authorAndy Polyakov <appro@openssl.org>
Fri, 13 Mar 2015 10:12:19 +0000 (11:12 +0100)
committerAndy Polyakov <appro@openssl.org>
Sat, 28 Mar 2015 14:59:27 +0000 (15:59 +0100)
Reviewed-by: Richard Levitte <levitte@openssl.org>
crypto/ec/asm/ecp_nistz256-x86_64.pl

index cdff22a..86ee382 100755 (executable)
 
 # Further optimization by <appro@openssl.org>:
 #
-#              this/original
-# Opteron      +12-49%
-# Bulldozer    +14-45%
-# P4           +18-46%
-# Westmere     +12-34%
-# Sandy Bridge +9-35%
-# Ivy Bridge   +9-35%
-# Haswell      +8-37%
-# Broadwell    +18-58%
-# Atom         +15-50%
-# VIA Nano     +43-160%
+#              this/original   with/without -DECP_NISTZ256_ASM(*)
+# Opteron      +12-49%         +110-150%
+# Bulldozer    +14-45%         +175-210%
+# P4           +18-46%         n/a :-(
+# Westmere     +12-34%         +80-87%
+# Sandy Bridge +9-35%          +110-120%
+# Ivy Bridge   +9-35%          +110-125%
+# Haswell      +8-37%          +140-160%
+# Broadwell    +18-58%         +145-210%
+# Atom         +15-50%         +130-180%
+# VIA Nano     +43-160%        +300-480%
+#
+# (*)  "without -DECP_NISTZ256_ASM" refers to build with
+#      "enable-ec_nistp_64_gcc_128";
 #
 # Ranges denote minimum and maximum improvement coefficients depending
-# on benchmark. Lower coefficients are for ECDSA sign, relatively
-# fastest server-side operation.
+# on benchmark. Lower coefficients are for ECDSA sign, relatively fastest
+# server-side operation. Keep in mind that +100% means 2x improvement.
 
 $flavour = shift;
 $output  = shift;