sha512-x86_64.pl: +15% better performance on Westmere and incidentally Atom.

Other Intel processors +5%, Opteron -2%.

1 files changed, 40 insertions, 46 deletions

diff --git a/crypto/sha/asm/sha512-x86_64.pl b/crypto/sha/asm/sha512-x86_64.pl
index e6643f8cf6..f611a2d898 100755
--- a/crypto/sha/asm/sha512-x86_64.pl
+++ b/crypto/sha/asm/sha512-x86_64.pl
@@ -95,50 +95,44 @@ sub ROUND_00_15()
 { my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
 
 $code.=<<___;
-	mov	$e,$a0
-	mov	$e,$a1
+	ror	\$`$Sigma1[2]-$Sigma1[1]`,$a0
 	mov	$f,$a2
+	mov	$T1,`$SZ*($i&0xf)`(%rsp)
 
-	ror	\$$Sigma1[0],$a0
-	ror	\$$Sigma1[1],$a1
+	ror	\$`$Sigma0[2]-$Sigma0[1]`,$a1
+	xor	$e,$a0
 	xor	$g,$a2			# f^g
 
-	xor	$a1,$a0
-	ror	\$`$Sigma1[2]-$Sigma1[1]`,$a1
+	ror	\$`$Sigma1[1]-$Sigma1[0]`,$a0
+	add	$h,$T1			# T1+=h
+	xor	$a,$a1
+
+	add	($Tbl,$round,$SZ),$T1	# T1+=K[round]
 	and	$e,$a2			# (f^g)&e
-	mov	$T1,`$SZ*($i&0xf)`(%rsp)
+	mov	$b,$h
 
-	xor	$a1,$a0			# Sigma1(e)
+	ror	\$`$Sigma0[1]-$Sigma0[0]`,$a1
+	xor	$e,$a0
 	xor	$g,$a2			# Ch(e,f,g)=((f^g)&e)^g
-	add	$h,$T1			# T1+=h
-
-	mov	$a,$h
-	add	$a0,$T1			# T1+=Sigma1(e)
 
+	xor	$c,$h			# b^c
+	xor	$a,$a1
 	add	$a2,$T1			# T1+=Ch(e,f,g)
-	mov	$a,$a0
-	mov	$a,$a1
+	mov	$b,$a2
 
-	ror	\$$Sigma0[0],$h
-	ror	\$$Sigma0[1],$a0
-	mov	$a,$a2
-	add	($Tbl,$round,$SZ),$T1	# T1+=K[round]
+	ror	\$$Sigma1[0],$a0	# Sigma1(e)
+	and	$a,$h			# h=(b^c)&a
+	and	$c,$a2			# b&c
 
-	xor	$a0,$h
-	ror	\$`$Sigma0[2]-$Sigma0[1]`,$a0
-	or	$c,$a1			# a|c
+	ror	\$$Sigma0[0],$a1	# Sigma0(a)
+	add	$a0,$T1			# T1+=Sigma1(e)
+	add	$a2,$h			# h+=b&c (completes +=Maj(a,b,c)
 
-	xor	$a0,$h			# h=Sigma0(a)
-	and	$c,$a2			# a&c
 	add	$T1,$d			# d+=T1
-
-	and	$b,$a1			# (a|c)&b
 	add	$T1,$h			# h+=T1
-
-	or	$a2,$a1			# Maj(a,b,c)=((a|c)&b)|(a&c)
 	lea	1($round),$round	# round++
+	add	$a1,$h			# h+=Sigma0(a)
 
-	add	$a1,$h			# h+=Maj(a,b,c)
 ___
 }
 
@@ -147,32 +141,30 @@ sub ROUND_16_XX()
 
 $code.=<<___;
 	mov	`$SZ*(($i+1)&0xf)`(%rsp),$a0
-	mov	`$SZ*(($i+14)&0xf)`(%rsp),$T1
-
-	mov	$a0,$a2
+	mov	`$SZ*(($i+14)&0xf)`(%rsp),$a1
+	mov	$a0,$T1
+	mov	$a1,$a2
 
+	ror	\$`$sigma0[1]-$sigma0[0]`,$T1
+	xor	$a0,$T1
 	shr	\$$sigma0[2],$a0
-	ror	\$$sigma0[0],$a2
-
-	xor	$a2,$a0
-	ror	\$`$sigma0[1]-$sigma0[0]`,$a2
 
-	xor	$a2,$a0			# sigma0(X[(i+1)&0xf])
-	mov	$T1,$a1
+	ror	\$$sigma0[0],$T1
+	xor	$T1,$a0			# sigma0(X[(i+1)&0xf])
+	mov	`$SZ*(($i+9)&0xf)`(%rsp),$T1
 
-	shr	\$$sigma1[2],$T1
-	ror	\$$sigma1[0],$a1
-
-	xor	$a1,$T1
-	ror	\$`$sigma1[1]-$sigma1[0]`,$a1
-
-	xor	$a1,$T1			# sigma1(X[(i+14)&0xf])
+	ror	\$`$sigma1[1]-$sigma1[0]`,$a2
+	xor	$a1,$a2
+	shr	\$$sigma1[2],$a1
 
+	ror	\$$sigma1[0],$a2
 	add	$a0,$T1
-
-	add	`$SZ*(($i+9)&0xf)`(%rsp),$T1
+	xor	$a2,$a1			# sigma1(X[(i+14)&0xf])
 
 	add	`$SZ*($i&0xf)`(%rsp),$T1
+	mov	$e,$a0
+	add	$a1,$T1
+	mov	$a,$a1
 ___
 	&ROUND_00_15(@_);
 }
@@ -219,6 +211,8 @@ $func:
 ___
 	for($i=0;$i<16;$i++) {
 		$code.="	mov	$SZ*$i($inp),$T1\n";
+		$code.="	mov	@ROT[4],$a0\n";
+		$code.="	mov	@ROT[0],$a1\n";
 		$code.="	bswap	$T1\n";
 		&ROUND_00_15($i,@ROT);
 		unshift(@ROT,pop(@ROT));