Merge branch 'develop'0.8.0

5 files changed, 798 insertions, 150 deletions

diff --git a/app/openssl/crypto/modes/asm/ghash-armv4.S b/app/openssl/crypto/modes/asm/ghash-armv4.S
index d66c4cbf..6c453774 100644
--- a/app/openssl/crypto/modes/asm/ghash-armv4.S
+++ b/app/openssl/crypto/modes/asm/ghash-armv4.S
@@ -309,99 +309,213 @@ gcm_gmult_4bit:
 #if __ARM_ARCH__>=7
 .fpu	neon
 
+.global	gcm_init_neon
+.type	gcm_init_neon,%function
+.align	4
+gcm_init_neon:
+	vld1.64		d7,[r1,:64]!	@ load H
+	vmov.i8		q8,#0xe1
+	vld1.64		d6,[r1,:64]
+	vshl.i64	d17,#57
+	vshr.u64	d16,#63		@ t0=0xc2....01
+	vdup.8		q9,d7[7]
+	vshr.u64	d26,d6,#63
+	vshr.s8		q9,#7			@ broadcast carry bit
+	vshl.i64	q3,q3,#1
+	vand		q8,q8,q9
+	vorr		d7,d26		@ H<<<=1
+	veor		q3,q3,q8		@ twisted H
+	vstmia		r0,{q3}
+
+	bx	lr					@ bx lr
+.size	gcm_init_neon,.-gcm_init_neon
+
 .global	gcm_gmult_neon
 .type	gcm_gmult_neon,%function
 .align	4
 gcm_gmult_neon:
-	sub		r1,#16		@ point at H in GCM128_CTX
-	vld1.64		d29,[r0,:64]!@ load Xi
-	vmov.i32	d5,#0xe1		@ our irreducible polynomial
-	vld1.64		d28,[r0,:64]!
-	vshr.u64	d5,#32
-	vldmia		r1,{d0-d1}	@ load H
-	veor		q12,q12
+	vld1.64		d7,[r0,:64]!	@ load Xi
+	vld1.64		d6,[r0,:64]!
+	vmov.i64	d29,#0x0000ffffffffffff
+	vldmia		r1,{d26-d27}	@ load twisted H
+	vmov.i64	d30,#0x00000000ffffffff
 #ifdef __ARMEL__
-	vrev64.8	q14,q14
+	vrev64.8	q3,q3
 #endif
-	veor		q13,q13
-	veor		q11,q11
-	mov		r1,#16
-	veor		q10,q10
+	vmov.i64	d31,#0x000000000000ffff
+	veor		d28,d26,d27		@ Karatsuba pre-processing
 	mov		r3,#16
-	veor		d2,d2
-	vdup.8		d4,d28[0]	@ broadcast lowest byte
-	b		.Linner_neon
+	b		.Lgmult_neon
 .size	gcm_gmult_neon,.-gcm_gmult_neon
 
 .global	gcm_ghash_neon
 .type	gcm_ghash_neon,%function
 .align	4
 gcm_ghash_neon:
-	vld1.64		d21,[r0,:64]!	@ load Xi
-	vmov.i32	d5,#0xe1		@ our irreducible polynomial
-	vld1.64		d20,[r0,:64]!
-	vshr.u64	d5,#32
-	vldmia		r0,{d0-d1}		@ load H
-	veor		q12,q12
-	nop
+	vld1.64		d1,[r0,:64]!	@ load Xi
+	vld1.64		d0,[r0,:64]!
+	vmov.i64	d29,#0x0000ffffffffffff
+	vldmia		r1,{d26-d27}	@ load twisted H
+	vmov.i64	d30,#0x00000000ffffffff
 #ifdef __ARMEL__
-	vrev64.8	q10,q10
+	vrev64.8	q0,q0
 #endif
-.Louter_neon:
-	vld1.64		d29,[r2]!	@ load inp
-	veor		q13,q13
-	vld1.64		d28,[r2]!
-	veor		q11,q11
-	mov		r1,#16
+	vmov.i64	d31,#0x000000000000ffff
+	veor		d28,d26,d27		@ Karatsuba pre-processing
+
+.Loop_neon:
+	vld1.64		d7,[r2]!		@ load inp
+	vld1.64		d6,[r2]!
 #ifdef __ARMEL__
-	vrev64.8	q14,q14
+	vrev64.8	q3,q3
 #endif
-	veor		d2,d2
-	veor		q14,q10			@ inp^=Xi
-	veor		q10,q10
-	vdup.8		d4,d28[0]	@ broadcast lowest byte
-.Linner_neon:
-	subs		r1,r1,#1
-	vmull.p8	q9,d1,d4		@ H.lo·Xi[i]
-	vmull.p8	q8,d0,d4		@ H.hi·Xi[i]
-	vext.8		q14,q12,#1		@ IN>>=8
-
-	veor		q10,q13		@ modulo-scheduled part
-	vshl.i64	d22,#48
-	vdup.8		d4,d28[0]	@ broadcast lowest byte
-	veor		d3,d18,d20
-
-	veor		d21,d22
-	vuzp.8		q9,q8
-	vsli.8		d2,d3,#1		@ compose the "carry" byte
-	vext.8		q10,q12,#1		@ Z>>=8
+	veor		q3,q0			@ inp^=Xi
+.Lgmult_neon:
+	vext.8		d16, d26, d26, #1	@ A1
+	vmull.p8	q8, d16, d6		@ F = A1*B
+	vext.8		d0, d6, d6, #1	@ B1
+	vmull.p8	q0, d26, d0		@ E = A*B1
+	vext.8		d18, d26, d26, #2	@ A2
+	vmull.p8	q9, d18, d6		@ H = A2*B
+	vext.8		d22, d6, d6, #2	@ B2
+	vmull.p8	q11, d26, d22		@ G = A*B2
+	vext.8		d20, d26, d26, #3	@ A3
+	veor		q8, q8, q0		@ L = E + F
+	vmull.p8	q10, d20, d6		@ J = A3*B
+	vext.8		d0, d6, d6, #3	@ B3
+	veor		q9, q9, q11		@ M = G + H
+	vmull.p8	q0, d26, d0		@ I = A*B3
+	veor		d16, d16, d17	@ t0 = (L) (P0 + P1) << 8
+	vand		d17, d17, d29
+	vext.8		d22, d6, d6, #4	@ B4
+	veor		d18, d18, d19	@ t1 = (M) (P2 + P3) << 16
+	vand		d19, d19, d30
+	vmull.p8	q11, d26, d22		@ K = A*B4
+	veor		q10, q10, q0		@ N = I + J
+	veor		d16, d16, d17
+	veor		d18, d18, d19
+	veor		d20, d20, d21	@ t2 = (N) (P4 + P5) << 24
+	vand		d21, d21, d31
+	vext.8		q8, q8, q8, #15
+	veor		d22, d22, d23	@ t3 = (K) (P6 + P7) << 32
+	vmov.i64	d23, #0
+	vext.8		q9, q9, q9, #14
+	veor		d20, d20, d21
+	vmull.p8	q0, d26, d6		@ D = A*B
+	vext.8		q11, q11, q11, #12
+	vext.8		q10, q10, q10, #13
+	veor		q8, q8, q9
+	veor		q10, q10, q11
+	veor		q0, q0, q8
+	veor		q0, q0, q10
+	veor		d6,d6,d7	@ Karatsuba pre-processing
+	vext.8		d16, d28, d28, #1	@ A1
+	vmull.p8	q8, d16, d6		@ F = A1*B
+	vext.8		d2, d6, d6, #1	@ B1
+	vmull.p8	q1, d28, d2		@ E = A*B1
+	vext.8		d18, d28, d28, #2	@ A2
+	vmull.p8	q9, d18, d6		@ H = A2*B
+	vext.8		d22, d6, d6, #2	@ B2
+	vmull.p8	q11, d28, d22		@ G = A*B2
+	vext.8		d20, d28, d28, #3	@ A3
+	veor		q8, q8, q1		@ L = E + F
+	vmull.p8	q10, d20, d6		@ J = A3*B
+	vext.8		d2, d6, d6, #3	@ B3
+	veor		q9, q9, q11		@ M = G + H
+	vmull.p8	q1, d28, d2		@ I = A*B3
+	veor		d16, d16, d17	@ t0 = (L) (P0 + P1) << 8
+	vand		d17, d17, d29
+	vext.8		d22, d6, d6, #4	@ B4
+	veor		d18, d18, d19	@ t1 = (M) (P2 + P3) << 16
+	vand		d19, d19, d30
+	vmull.p8	q11, d28, d22		@ K = A*B4
+	veor		q10, q10, q1		@ N = I + J
+	veor		d16, d16, d17
+	veor		d18, d18, d19
+	veor		d20, d20, d21	@ t2 = (N) (P4 + P5) << 24
+	vand		d21, d21, d31
+	vext.8		q8, q8, q8, #15
+	veor		d22, d22, d23	@ t3 = (K) (P6 + P7) << 32
+	vmov.i64	d23, #0
+	vext.8		q9, q9, q9, #14
+	veor		d20, d20, d21
+	vmull.p8	q1, d28, d6		@ D = A*B
+	vext.8		q11, q11, q11, #12
+	vext.8		q10, q10, q10, #13
+	veor		q8, q8, q9
+	veor		q10, q10, q11
+	veor		q1, q1, q8
+	veor		q1, q1, q10
+	vext.8		d16, d27, d27, #1	@ A1
+	vmull.p8	q8, d16, d7		@ F = A1*B
+	vext.8		d4, d7, d7, #1	@ B1
+	vmull.p8	q2, d27, d4		@ E = A*B1
+	vext.8		d18, d27, d27, #2	@ A2
+	vmull.p8	q9, d18, d7		@ H = A2*B
+	vext.8		d22, d7, d7, #2	@ B2
+	vmull.p8	q11, d27, d22		@ G = A*B2
+	vext.8		d20, d27, d27, #3	@ A3
+	veor		q8, q8, q2		@ L = E + F
+	vmull.p8	q10, d20, d7		@ J = A3*B
+	vext.8		d4, d7, d7, #3	@ B3
+	veor		q9, q9, q11		@ M = G + H
+	vmull.p8	q2, d27, d4		@ I = A*B3
+	veor		d16, d16, d17	@ t0 = (L) (P0 + P1) << 8
+	vand		d17, d17, d29
+	vext.8		d22, d7, d7, #4	@ B4
+	veor		d18, d18, d19	@ t1 = (M) (P2 + P3) << 16
+	vand		d19, d19, d30
+	vmull.p8	q11, d27, d22		@ K = A*B4
+	veor		q10, q10, q2		@ N = I + J
+	veor		d16, d16, d17
+	veor		d18, d18, d19
+	veor		d20, d20, d21	@ t2 = (N) (P4 + P5) << 24
+	vand		d21, d21, d31
+	vext.8		q8, q8, q8, #15
+	veor		d22, d22, d23	@ t3 = (K) (P6 + P7) << 32
+	vmov.i64	d23, #0
+	vext.8		q9, q9, q9, #14
+	veor		d20, d20, d21
+	vmull.p8	q2, d27, d7		@ D = A*B
+	vext.8		q11, q11, q11, #12
+	vext.8		q10, q10, q10, #13
+	veor		q8, q8, q9
+	veor		q10, q10, q11
+	veor		q2, q2, q8
+	veor		q2, q2, q10
+	veor		q1,q1,q0		@ Karatsuba post-processing
+	veor		q1,q1,q2
+	veor		d1,d1,d2
+	veor		d4,d4,d3	@ Xh|Xl - 256-bit result
 
-	vmull.p8	q11,d2,d5		@ "carry"·0xe1
-	vshr.u8		d2,d3,#7		@ save Z's bottom bit
-	vext.8		q13,q9,q12,#1	@ Qlo>>=8
-	veor		q10,q8
-	bne		.Linner_neon
+	@ equivalent of reduction_avx from ghash-x86_64.pl
+	vshl.i64	q9,q0,#57		@ 1st phase
+	vshl.i64	q10,q0,#62
+	veor		q10,q10,q9		@
+	vshl.i64	q9,q0,#63
+	veor		q10, q10, q9		@
+ 	veor		d1,d1,d20	@
+	veor		d4,d4,d21
 
-	veor		q10,q13		@ modulo-scheduled artefact
-	vshl.i64	d22,#48
-	veor		d21,d22
+	vshr.u64	q10,q0,#1		@ 2nd phase
+	veor		q2,q2,q0
+	veor		q0,q0,q10		@
+	vshr.u64	q10,q10,#6
+	vshr.u64	q0,q0,#1		@
+	veor		q0,q0,q2		@
+	veor		q0,q0,q10		@
 
-	@ finalization, normalize Z:Zo
-	vand		d2,d5		@ suffices to mask the bit
-	vshr.u64	d3,d20,#63
-	vshl.i64	q10,#1
 	subs		r3,#16
-	vorr		q10,q1		@ Z=Z:Zo<<1
-	bne		.Louter_neon
+	bne		.Loop_neon
 
 #ifdef __ARMEL__
-	vrev64.8	q10,q10
+	vrev64.8	q0,q0
 #endif
 	sub		r0,#16	
-	vst1.64		d21,[r0,:64]!	@ write out Xi
-	vst1.64		d20,[r0,:64]
+	vst1.64		d1,[r0,:64]!	@ write out Xi
+	vst1.64		d0,[r0,:64]
 
-	.word	0xe12fff1e
+	bx	lr					@ bx lr
 .size	gcm_ghash_neon,.-gcm_ghash_neon
 #endif
 .asciz  "GHASH for ARMv4/NEON, CRYPTOGAMS by <appro@openssl.org>"
diff --git a/app/openssl/crypto/modes/asm/ghash-armv4.pl b/app/openssl/crypto/modes/asm/ghash-armv4.pl
index e46f8e34..b79ecbcc 100644
--- a/app/openssl/crypto/modes/asm/ghash-armv4.pl
+++ b/app/openssl/crypto/modes/asm/ghash-armv4.pl
@@ -35,6 +35,20 @@
 # Add NEON implementation featuring polynomial multiplication, i.e. no
 # lookup tables involved. On Cortex A8 it was measured to process one
 # byte in 15 cycles or 55% faster than integer-only code.
+#
+# April 2014
+#
+# Switch to multiplication algorithm suggested in paper referred
+# below and combine it with reduction algorithm from x86 module.
+# Performance improvement over previous version varies from 65% on
+# Snapdragon S4 to 110% on Cortex A9. In absolute terms Cortex A8
+# processes one byte in 8.45 cycles, A9 - in 10.2, Snapdragon S4 -
+# in 9.33.
+#
+# Câmara, D.; Gouvêa, C. P. L.; López, J. & Dahab, R.: Fast Software
+# Polynomial Multiplication on ARM Processors using the NEON Engine.
+# 
+# http://conradoplg.cryptoland.net/files/2010/12/mocrysen13.pdf
 
 # ====================================================================
 # Note about "528B" variant. In ARM case it makes lesser sense to
@@ -303,117 +317,160 @@ $code.=<<___;
 .size	gcm_gmult_4bit,.-gcm_gmult_4bit
 ___
 {
-my $cnt=$Htbl;	# $Htbl is used once in the very beginning
-
-my ($Hhi, $Hlo, $Zo, $T, $xi, $mod) = map("d$_",(0..7));
-my ($Qhi, $Qlo, $Z,  $R, $zero, $Qpost, $IN) = map("q$_",(8..15));
-
-# Z:Zo keeps 128-bit result shifted by 1 to the right, with bottom bit
-# in Zo. Or should I say "top bit", because GHASH is specified in
-# reverse bit order? Otherwise straightforward 128-bt H by one input
-# byte multiplication and modulo-reduction, times 16.
+my ($Xl,$Xm,$Xh,$IN)=map("q$_",(0..3));
+my ($t0,$t1,$t2,$t3)=map("q$_",(8..12));
+my ($Hlo,$Hhi,$Hhl,$k48,$k32,$k16)=map("d$_",(26..31));
 
-sub Dlo()   { shift=~m|q([1]?[0-9])|?"d".($1*2):"";     }
-sub Dhi()   { shift=~m|q([1]?[0-9])|?"d".($1*2+1):"";   }
-sub Q()     { shift=~m|d([1-3]?[02468])|?"q".($1/2):""; }
+sub clmul64x64 {
+my ($r,$a,$b)=@_;
+$code.=<<___;
+	vext.8		$t0#lo, $a, $a, #1	@ A1
+	vmull.p8	$t0, $t0#lo, $b		@ F = A1*B
+	vext.8		$r#lo, $b, $b, #1	@ B1
+	vmull.p8	$r, $a, $r#lo		@ E = A*B1
+	vext.8		$t1#lo, $a, $a, #2	@ A2
+	vmull.p8	$t1, $t1#lo, $b		@ H = A2*B
+	vext.8		$t3#lo, $b, $b, #2	@ B2
+	vmull.p8	$t3, $a, $t3#lo		@ G = A*B2
+	vext.8		$t2#lo, $a, $a, #3	@ A3
+	veor		$t0, $t0, $r		@ L = E + F
+	vmull.p8	$t2, $t2#lo, $b		@ J = A3*B
+	vext.8		$r#lo, $b, $b, #3	@ B3
+	veor		$t1, $t1, $t3		@ M = G + H
+	vmull.p8	$r, $a, $r#lo		@ I = A*B3
+	veor		$t0#lo, $t0#lo, $t0#hi	@ t0 = (L) (P0 + P1) << 8
+	vand		$t0#hi, $t0#hi, $k48
+	vext.8		$t3#lo, $b, $b, #4	@ B4
+	veor		$t1#lo, $t1#lo, $t1#hi	@ t1 = (M) (P2 + P3) << 16
+	vand		$t1#hi, $t1#hi, $k32
+	vmull.p8	$t3, $a, $t3#lo		@ K = A*B4
+	veor		$t2, $t2, $r		@ N = I + J
+	veor		$t0#lo, $t0#lo, $t0#hi
+	veor		$t1#lo, $t1#lo, $t1#hi
+	veor		$t2#lo, $t2#lo, $t2#hi	@ t2 = (N) (P4 + P5) << 24
+	vand		$t2#hi, $t2#hi, $k16
+	vext.8		$t0, $t0, $t0, #15
+	veor		$t3#lo, $t3#lo, $t3#hi	@ t3 = (K) (P6 + P7) << 32
+	vmov.i64	$t3#hi, #0
+	vext.8		$t1, $t1, $t1, #14
+	veor		$t2#lo, $t2#lo, $t2#hi
+	vmull.p8	$r, $a, $b		@ D = A*B
+	vext.8		$t3, $t3, $t3, #12
+	vext.8		$t2, $t2, $t2, #13
+	veor		$t0, $t0, $t1
+	veor		$t2, $t2, $t3
+	veor		$r, $r, $t0
+	veor		$r, $r, $t2
+___
+}
 
 $code.=<<___;
 #if __ARM_ARCH__>=7
 .fpu	neon
 
+.global	gcm_init_neon
+.type	gcm_init_neon,%function
+.align	4
+gcm_init_neon:
+	vld1.64		$IN#hi,[r1,:64]!	@ load H
+	vmov.i8		$t0,#0xe1
+	vld1.64		$IN#lo,[r1,:64]
+	vshl.i64	$t0#hi,#57
+	vshr.u64	$t0#lo,#63		@ t0=0xc2....01
+	vdup.8		$t1,$IN#hi[7]
+	vshr.u64	$Hlo,$IN#lo,#63
+	vshr.s8		$t1,#7			@ broadcast carry bit
+	vshl.i64	$IN,$IN,#1
+	vand		$t0,$t0,$t1
+	vorr		$IN#hi,$Hlo		@ H<<<=1
+	veor		$IN,$IN,$t0		@ twisted H
+	vstmia		r0,{$IN}
+
+	ret					@ bx lr
+.size	gcm_init_neon,.-gcm_init_neon
+
 .global	gcm_gmult_neon
 .type	gcm_gmult_neon,%function
 .align	4
 gcm_gmult_neon:
-	sub		$Htbl,#16		@ point at H in GCM128_CTX
-	vld1.64		`&Dhi("$IN")`,[$Xi,:64]!@ load Xi
-	vmov.i32	$mod,#0xe1		@ our irreducible polynomial
-	vld1.64		`&Dlo("$IN")`,[$Xi,:64]!
-	vshr.u64	$mod,#32
-	vldmia		$Htbl,{$Hhi-$Hlo}	@ load H
-	veor		$zero,$zero
+	vld1.64		$IN#hi,[$Xi,:64]!	@ load Xi
+	vld1.64		$IN#lo,[$Xi,:64]!
+	vmov.i64	$k48,#0x0000ffffffffffff
+	vldmia		$Htbl,{$Hlo-$Hhi}	@ load twisted H
+	vmov.i64	$k32,#0x00000000ffffffff
 #ifdef __ARMEL__
 	vrev64.8	$IN,$IN
 #endif
-	veor		$Qpost,$Qpost
-	veor		$R,$R
-	mov		$cnt,#16
-	veor		$Z,$Z
+	vmov.i64	$k16,#0x000000000000ffff
+	veor		$Hhl,$Hlo,$Hhi		@ Karatsuba pre-processing
 	mov		$len,#16
-	veor		$Zo,$Zo
-	vdup.8		$xi,`&Dlo("$IN")`[0]	@ broadcast lowest byte
-	b		.Linner_neon
+	b		.Lgmult_neon
 .size	gcm_gmult_neon,.-gcm_gmult_neon
 
 .global	gcm_ghash_neon
 .type	gcm_ghash_neon,%function
 .align	4
 gcm_ghash_neon:
-	vld1.64		`&Dhi("$Z")`,[$Xi,:64]!	@ load Xi
-	vmov.i32	$mod,#0xe1		@ our irreducible polynomial
-	vld1.64		`&Dlo("$Z")`,[$Xi,:64]!
-	vshr.u64	$mod,#32
-	vldmia		$Xi,{$Hhi-$Hlo}		@ load H
-	veor		$zero,$zero
-	nop
+	vld1.64		$Xl#hi,[$Xi,:64]!	@ load Xi
+	vld1.64		$Xl#lo,[$Xi,:64]!
+	vmov.i64	$k48,#0x0000ffffffffffff
+	vldmia		$Htbl,{$Hlo-$Hhi}	@ load twisted H
+	vmov.i64	$k32,#0x00000000ffffffff
 #ifdef __ARMEL__
-	vrev64.8	$Z,$Z
+	vrev64.8	$Xl,$Xl
 #endif
-.Louter_neon:
-	vld1.64		`&Dhi($IN)`,[$inp]!	@ load inp
-	veor		$Qpost,$Qpost
-	vld1.64		`&Dlo($IN)`,[$inp]!
-	veor		$R,$R
-	mov		$cnt,#16
+	vmov.i64	$k16,#0x000000000000ffff
+	veor		$Hhl,$Hlo,$Hhi		@ Karatsuba pre-processing
+
+.Loop_neon:
+	vld1.64		$IN#hi,[$inp]!		@ load inp
+	vld1.64		$IN#lo,[$inp]!
 #ifdef __ARMEL__
 	vrev64.8	$IN,$IN
 #endif
-	veor		$Zo,$Zo
-	veor		$IN,$Z			@ inp^=Xi
-	veor		$Z,$Z
-	vdup.8		$xi,`&Dlo("$IN")`[0]	@ broadcast lowest byte
-.Linner_neon:
-	subs		$cnt,$cnt,#1
-	vmull.p8	$Qlo,$Hlo,$xi		@ H.lo·Xi[i]
-	vmull.p8	$Qhi,$Hhi,$xi		@ H.hi·Xi[i]
-	vext.8		$IN,$zero,#1		@ IN>>=8
-
-	veor		$Z,$Qpost		@ modulo-scheduled part
-	vshl.i64	`&Dlo("$R")`,#48
-	vdup.8		$xi,`&Dlo("$IN")`[0]	@ broadcast lowest byte
-	veor		$T,`&Dlo("$Qlo")`,`&Dlo("$Z")`
-
-	veor		`&Dhi("$Z")`,`&Dlo("$R")`
-	vuzp.8		$Qlo,$Qhi
-	vsli.8		$Zo,$T,#1		@ compose the "carry" byte
-	vext.8		$Z,$zero,#1		@ Z>>=8
-
-	vmull.p8	$R,$Zo,$mod		@ "carry"·0xe1
-	vshr.u8		$Zo,$T,#7		@ save Z's bottom bit
-	vext.8		$Qpost,$Qlo,$zero,#1	@ Qlo>>=8
-	veor		$Z,$Qhi
-	bne		.Linner_neon
-
-	veor		$Z,$Qpost		@ modulo-scheduled artefact
-	vshl.i64	`&Dlo("$R")`,#48
-	veor		`&Dhi("$Z")`,`&Dlo("$R")`
-
-	@ finalization, normalize Z:Zo
-	vand		$Zo,$mod		@ suffices to mask the bit
-	vshr.u64	`&Dhi(&Q("$Zo"))`,`&Dlo("$Z")`,#63
-	vshl.i64	$Z,#1
+	veor		$IN,$Xl			@ inp^=Xi
+.Lgmult_neon:
+___
+	&clmul64x64	($Xl,$Hlo,"$IN#lo");	# H.lo·Xi.lo
+$code.=<<___;
+	veor		$IN#lo,$IN#lo,$IN#hi	@ Karatsuba pre-processing
+___
+	&clmul64x64	($Xm,$Hhl,"$IN#lo");	# (H.lo+H.hi)·(Xi.lo+Xi.hi)
+	&clmul64x64	($Xh,$Hhi,"$IN#hi");	# H.hi·Xi.hi
+$code.=<<___;
+	veor		$Xm,$Xm,$Xl		@ Karatsuba post-processing
+	veor		$Xm,$Xm,$Xh
+	veor		$Xl#hi,$Xl#hi,$Xm#lo
+	veor		$Xh#lo,$Xh#lo,$Xm#hi	@ Xh|Xl - 256-bit result
+
+	@ equivalent of reduction_avx from ghash-x86_64.pl
+	vshl.i64	$t1,$Xl,#57		@ 1st phase
+	vshl.i64	$t2,$Xl,#62
+	veor		$t2,$t2,$t1		@
+	vshl.i64	$t1,$Xl,#63
+	veor		$t2, $t2, $t1		@
+ 	veor		$Xl#hi,$Xl#hi,$t2#lo	@
+	veor		$Xh#lo,$Xh#lo,$t2#hi
+
+	vshr.u64	$t2,$Xl,#1		@ 2nd phase
+	veor		$Xh,$Xh,$Xl
+	veor		$Xl,$Xl,$t2		@
+	vshr.u64	$t2,$t2,#6
+	vshr.u64	$Xl,$Xl,#1		@
+	veor		$Xl,$Xl,$Xh		@
+	veor		$Xl,$Xl,$t2		@
+
 	subs		$len,#16
-	vorr		$Z,`&Q("$Zo")`		@ Z=Z:Zo<<1
-	bne		.Louter_neon
+	bne		.Loop_neon
 
 #ifdef __ARMEL__
-	vrev64.8	$Z,$Z
+	vrev64.8	$Xl,$Xl
 #endif
 	sub		$Xi,#16	
-	vst1.64		`&Dhi("$Z")`,[$Xi,:64]!	@ write out Xi
-	vst1.64		`&Dlo("$Z")`,[$Xi,:64]
+	vst1.64		$Xl#hi,[$Xi,:64]!	@ write out Xi
+	vst1.64		$Xl#lo,[$Xi,:64]
 
-	bx	lr
+	ret					@ bx lr
 .size	gcm_ghash_neon,.-gcm_ghash_neon
 #endif
 ___
@@ -423,7 +480,13 @@ $code.=<<___;
 .align  2
 ___
 
-$code =~ s/\`([^\`]*)\`/eval $1/gem;
-$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
-print $code;
+foreach (split("\n",$code)) {
+	s/\`([^\`]*)\`/eval $1/geo;
+
+	s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo	or
+	s/\bret\b/bx	lr/go		or
+	s/\bbx\s+lr\b/.word\t0xe12fff1e/go;    # make it possible to compile with -march=armv4
+
+	print $_,"\n";
+}
 close STDOUT; # enforce flush
diff --git a/app/openssl/crypto/modes/asm/ghashv8-armx-64.S b/app/openssl/crypto/modes/asm/ghashv8-armx-64.S
new file mode 100644
index 00000000..b77b6c40
--- /dev/null
+++ b/app/openssl/crypto/modes/asm/ghashv8-armx-64.S
@@ -0,0 +1,115 @@
+#include "arm_arch.h"
+
+.text
+.arch	armv8-a+crypto
+.global	gcm_init_v8
+.type	gcm_init_v8,%function
+.align	4
+gcm_init_v8:
+	ld1		{v17.2d},[x1]		//load H
+	movi		v16.16b,#0xe1
+	ext		v3.16b,v17.16b,v17.16b,#8
+	shl	v16.2d,v16.2d,#57
+	ushr	v18.2d,v16.2d,#63
+	ext		v16.16b,v18.16b,v16.16b,#8		//t0=0xc2....01
+	dup		v17.4s,v17.s[1]
+	ushr	v19.2d,v3.2d,#63
+	sshr	v17.4s,v17.4s,#31		//broadcast carry bit
+	and		v19.16b,v19.16b,v16.16b
+	shl	v3.2d,v3.2d,#1
+	ext		v19.16b,v19.16b,v19.16b,#8
+	and		v16.16b,v16.16b,v17.16b
+	orr		v3.16b,v3.16b,v19.16b		//H<<<=1
+	eor		v3.16b,v3.16b,v16.16b		//twisted H
+	st1		{v3.2d},[x0]
+
+	ret
+.size	gcm_init_v8,.-gcm_init_v8
+
+.global	gcm_gmult_v8
+.type	gcm_gmult_v8,%function
+.align	4
+gcm_gmult_v8:
+	ld1		{v17.2d},[x0]		//load Xi
+	movi		v19.16b,#0xe1
+	ld1		{v20.2d},[x1]		//load twisted H
+	shl	v19.2d,v19.2d,#57
+#ifndef __ARMEB__
+	rev64	v17.16b,v17.16b
+#endif
+	ext		v21.16b,v20.16b,v20.16b,#8
+	mov		x3,#0
+	ext		v3.16b,v17.16b,v17.16b,#8
+	mov		x12,#0
+	eor		v21.16b,v21.16b,v20.16b		//Karatsuba pre-processing
+	mov		x2,x0
+	b		.Lgmult_v8
+.size	gcm_gmult_v8,.-gcm_gmult_v8
+
+.global	gcm_ghash_v8
+.type	gcm_ghash_v8,%function
+.align	4
+gcm_ghash_v8:
+	ld1		{v0.2d},[x0]		//load [rotated] Xi
+	subs		x3,x3,#16
+	movi		v19.16b,#0xe1
+	mov		x12,#16
+	ld1		{v20.2d},[x1]		//load twisted H
+	csel	x12,xzr,x12,eq
+	ext		v0.16b,v0.16b,v0.16b,#8
+	shl	v19.2d,v19.2d,#57
+	ld1		{v17.2d},[x2],x12	//load [rotated] inp
+	ext		v21.16b,v20.16b,v20.16b,#8
+#ifndef __ARMEB__
+	rev64	v0.16b,v0.16b
+	rev64	v17.16b,v17.16b
+#endif
+	eor		v21.16b,v21.16b,v20.16b		//Karatsuba pre-processing
+	ext		v3.16b,v17.16b,v17.16b,#8
+	b		.Loop_v8
+
+.align	4
+.Loop_v8:
+	ext		v18.16b,v0.16b,v0.16b,#8
+	eor		v3.16b,v3.16b,v0.16b		//inp^=Xi
+	eor		v17.16b,v17.16b,v18.16b		//v17.16b is rotated inp^Xi
+
+.Lgmult_v8:
+	pmull	v0.1q,v20.1d,v3.1d		//H.lo·Xi.lo
+	eor		v17.16b,v17.16b,v3.16b		//Karatsuba pre-processing
+	pmull2	v2.1q,v20.2d,v3.2d		//H.hi·Xi.hi
+	subs		x3,x3,#16
+	pmull	v1.1q,v21.1d,v17.1d		//(H.lo+H.hi)·(Xi.lo+Xi.hi)
+	csel	x12,xzr,x12,eq
+
+	ext		v17.16b,v0.16b,v2.16b,#8		//Karatsuba post-processing
+	eor		v18.16b,v0.16b,v2.16b
+	eor		v1.16b,v1.16b,v17.16b
+	 ld1	{v17.2d},[x2],x12	//load [rotated] inp
+	eor		v1.16b,v1.16b,v18.16b
+	pmull	v18.1q,v0.1d,v19.1d		//1st phase
+
+	ins	v2.d[0],v1.d[1]
+	ins	v1.d[1],v0.d[0]
+#ifndef __ARMEB__
+	 rev64	v17.16b,v17.16b
+#endif
+	eor		v0.16b,v1.16b,v18.16b
+	 ext		v3.16b,v17.16b,v17.16b,#8
+
+	ext		v18.16b,v0.16b,v0.16b,#8		//2nd phase
+	pmull	v0.1q,v0.1d,v19.1d
+	eor		v18.16b,v18.16b,v2.16b
+	eor		v0.16b,v0.16b,v18.16b
+	b.hs		.Loop_v8
+
+#ifndef __ARMEB__
+	rev64	v0.16b,v0.16b
+#endif
+	ext		v0.16b,v0.16b,v0.16b,#8
+	st1		{v0.2d},[x0]		//write out Xi
+
+	ret
+.size	gcm_ghash_v8,.-gcm_ghash_v8
+.asciz  "GHASH for ARMv8, CRYPTOGAMS by <appro@openssl.org>"
+.align  2
diff --git a/app/openssl/crypto/modes/asm/ghashv8-armx.S b/app/openssl/crypto/modes/asm/ghashv8-armx.S
new file mode 100644
index 00000000..f388c54e
--- /dev/null
+++ b/app/openssl/crypto/modes/asm/ghashv8-armx.S
@@ -0,0 +1,116 @@
+#include "arm_arch.h"
+
+.text
+.fpu	neon
+.code	32
+.global	gcm_init_v8
+.type	gcm_init_v8,%function
+.align	4
+gcm_init_v8:
+	vld1.64		{q9},[r1]		@ load H
+	vmov.i8		q8,#0xe1
+	vext.8		q3,q9,q9,#8
+	vshl.i64	q8,q8,#57
+	vshr.u64	q10,q8,#63
+	vext.8		q8,q10,q8,#8		@ t0=0xc2....01
+	vdup.32	q9,d18[1]
+	vshr.u64	q11,q3,#63
+	vshr.s32	q9,q9,#31		@ broadcast carry bit
+	vand		q11,q11,q8
+	vshl.i64	q3,q3,#1
+	vext.8		q11,q11,q11,#8
+	vand		q8,q8,q9
+	vorr		q3,q3,q11		@ H<<<=1
+	veor		q3,q3,q8		@ twisted H
+	vst1.64		{q3},[r0]
+
+	bx	lr
+.size	gcm_init_v8,.-gcm_init_v8
+
+.global	gcm_gmult_v8
+.type	gcm_gmult_v8,%function
+.align	4
+gcm_gmult_v8:
+	vld1.64		{q9},[r0]		@ load Xi
+	vmov.i8		q11,#0xe1
+	vld1.64		{q12},[r1]		@ load twisted H
+	vshl.u64	q11,q11,#57
+#ifndef __ARMEB__
+	vrev64.8	q9,q9
+#endif
+	vext.8		q13,q12,q12,#8
+	mov		r3,#0
+	vext.8		q3,q9,q9,#8
+	mov		r12,#0
+	veor		q13,q13,q12		@ Karatsuba pre-processing
+	mov		r2,r0
+	b		.Lgmult_v8
+.size	gcm_gmult_v8,.-gcm_gmult_v8
+
+.global	gcm_ghash_v8
+.type	gcm_ghash_v8,%function
+.align	4
+gcm_ghash_v8:
+	vld1.64		{q0},[r0]		@ load [rotated] Xi
+	subs		r3,r3,#16
+	vmov.i8		q11,#0xe1
+	mov		r12,#16
+	vld1.64		{q12},[r1]		@ load twisted H
+	moveq	r12,#0
+	vext.8		q0,q0,q0,#8
+	vshl.u64	q11,q11,#57
+	vld1.64		{q9},[r2],r12	@ load [rotated] inp
+	vext.8		q13,q12,q12,#8
+#ifndef __ARMEB__
+	vrev64.8	q0,q0
+	vrev64.8	q9,q9
+#endif
+	veor		q13,q13,q12		@ Karatsuba pre-processing
+	vext.8		q3,q9,q9,#8
+	b		.Loop_v8
+
+.align	4
+.Loop_v8:
+	vext.8		q10,q0,q0,#8
+	veor		q3,q3,q0		@ inp^=Xi
+	veor		q9,q9,q10		@ q9 is rotated inp^Xi
+
+.Lgmult_v8:
+	.byte	0x86,0x0e,0xa8,0xf2	@ pmull q0,q12,q3		@ H.lo·Xi.lo
+	veor		q9,q9,q3		@ Karatsuba pre-processing
+	.byte	0x87,0x4e,0xa9,0xf2	@ pmull2 q2,q12,q3		@ H.hi·Xi.hi
+	subs		r3,r3,#16
+	.byte	0xa2,0x2e,0xaa,0xf2	@ pmull q1,q13,q9		@ (H.lo+H.hi)·(Xi.lo+Xi.hi)
+	moveq	r12,#0
+
+	vext.8		q9,q0,q2,#8		@ Karatsuba post-processing
+	veor		q10,q0,q2
+	veor		q1,q1,q9
+	 vld1.64	{q9},[r2],r12	@ load [rotated] inp
+	veor		q1,q1,q10
+	.byte	0x26,0x4e,0xe0,0xf2	@ pmull q10,q0,q11		@ 1st phase
+
+	vmov		d4,d3		@ Xh|Xm - 256-bit result
+	vmov		d3,d0		@ Xm is rotated Xl
+#ifndef __ARMEB__
+	 vrev64.8	q9,q9
+#endif
+	veor		q0,q1,q10
+	 vext.8		q3,q9,q9,#8
+
+	vext.8		q10,q0,q0,#8		@ 2nd phase
+	.byte	0x26,0x0e,0xa0,0xf2	@ pmull q0,q0,q11
+	veor		q10,q10,q2
+	veor		q0,q0,q10
+	bhs		.Loop_v8
+
+#ifndef __ARMEB__
+	vrev64.8	q0,q0
+#endif
+	vext.8		q0,q0,q0,#8
+	vst1.64		{q0},[r0]		@ write out Xi
+
+	bx	lr
+.size	gcm_ghash_v8,.-gcm_ghash_v8
+.asciz  "GHASH for ARMv8, CRYPTOGAMS by <appro@openssl.org>"
+.align  2
diff --git a/app/openssl/crypto/modes/asm/ghashv8-armx.pl b/app/openssl/crypto/modes/asm/ghashv8-armx.pl
new file mode 100644
index 00000000..69e863e7
--- /dev/null
+++ b/app/openssl/crypto/modes/asm/ghashv8-armx.pl
@@ -0,0 +1,240 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# GHASH for ARMv8 Crypto Extension, 64-bit polynomial multiplication.
+#
+# June 2014
+#
+# Initial version was developed in tight cooperation with Ard
+# Biesheuvel <ard.biesheuvel@linaro.org> from bits-n-pieces from
+# other assembly modules. Just like aesv8-armx.pl this module
+# supports both AArch32 and AArch64 execution modes.
+#
+# Current performance in cycles per processed byte:
+#
+#		PMULL[2]	32-bit NEON(*)
+# Apple A7	1.76		5.62
+# Cortex-A5x	n/a		n/a
+#
+# (*)	presented for reference/comparison purposes;
+
+$flavour = shift;
+open STDOUT,">".shift;
+
+$Xi="x0";	# argument block
+$Htbl="x1";
+$inp="x2";
+$len="x3";
+
+$inc="x12";
+
+{
+my ($Xl,$Xm,$Xh,$IN)=map("q$_",(0..3));
+my ($t0,$t1,$t2,$t3,$H,$Hhl)=map("q$_",(8..14));
+
+$code=<<___;
+#include "arm_arch.h"
+
+.text
+___
+$code.=".arch	armv8-a+crypto\n"	if ($flavour =~ /64/);
+$code.=".fpu	neon\n.code	32\n"	if ($flavour !~ /64/);
+
+$code.=<<___;
+.global	gcm_init_v8
+.type	gcm_init_v8,%function
+.align	4
+gcm_init_v8:
+	vld1.64		{$t1},[x1]		@ load H
+	vmov.i8		$t0,#0xe1
+	vext.8		$IN,$t1,$t1,#8
+	vshl.i64	$t0,$t0,#57
+	vshr.u64	$t2,$t0,#63
+	vext.8		$t0,$t2,$t0,#8		@ t0=0xc2....01
+	vdup.32		$t1,${t1}[1]
+	vshr.u64	$t3,$IN,#63
+	vshr.s32	$t1,$t1,#31		@ broadcast carry bit
+	vand		$t3,$t3,$t0
+	vshl.i64	$IN,$IN,#1
+	vext.8		$t3,$t3,$t3,#8
+	vand		$t0,$t0,$t1
+	vorr		$IN,$IN,$t3		@ H<<<=1
+	veor		$IN,$IN,$t0		@ twisted H
+	vst1.64		{$IN},[x0]
+
+	ret
+.size	gcm_init_v8,.-gcm_init_v8
+
+.global	gcm_gmult_v8
+.type	gcm_gmult_v8,%function
+.align	4
+gcm_gmult_v8:
+	vld1.64		{$t1},[$Xi]		@ load Xi
+	vmov.i8		$t3,#0xe1
+	vld1.64		{$H},[$Htbl]		@ load twisted H
+	vshl.u64	$t3,$t3,#57
+#ifndef __ARMEB__
+	vrev64.8	$t1,$t1
+#endif
+	vext.8		$Hhl,$H,$H,#8
+	mov		$len,#0
+	vext.8		$IN,$t1,$t1,#8
+	mov		$inc,#0
+	veor		$Hhl,$Hhl,$H		@ Karatsuba pre-processing
+	mov		$inp,$Xi
+	b		.Lgmult_v8
+.size	gcm_gmult_v8,.-gcm_gmult_v8
+
+.global	gcm_ghash_v8
+.type	gcm_ghash_v8,%function
+.align	4
+gcm_ghash_v8:
+	vld1.64		{$Xl},[$Xi]		@ load [rotated] Xi
+	subs		$len,$len,#16
+	vmov.i8		$t3,#0xe1
+	mov		$inc,#16
+	vld1.64		{$H},[$Htbl]		@ load twisted H
+	cclr		$inc,eq
+	vext.8		$Xl,$Xl,$Xl,#8
+	vshl.u64	$t3,$t3,#57
+	vld1.64		{$t1},[$inp],$inc	@ load [rotated] inp
+	vext.8		$Hhl,$H,$H,#8
+#ifndef __ARMEB__
+	vrev64.8	$Xl,$Xl
+	vrev64.8	$t1,$t1
+#endif
+	veor		$Hhl,$Hhl,$H		@ Karatsuba pre-processing
+	vext.8		$IN,$t1,$t1,#8
+	b		.Loop_v8
+
+.align	4
+.Loop_v8:
+	vext.8		$t2,$Xl,$Xl,#8
+	veor		$IN,$IN,$Xl		@ inp^=Xi
+	veor		$t1,$t1,$t2		@ $t1 is rotated inp^Xi
+
+.Lgmult_v8:
+	vpmull.p64	$Xl,$H,$IN		@ H.lo·Xi.lo
+	veor		$t1,$t1,$IN		@ Karatsuba pre-processing
+	vpmull2.p64	$Xh,$H,$IN		@ H.hi·Xi.hi
+	subs		$len,$len,#16
+	vpmull.p64	$Xm,$Hhl,$t1		@ (H.lo+H.hi)·(Xi.lo+Xi.hi)
+	cclr		$inc,eq
+
+	vext.8		$t1,$Xl,$Xh,#8		@ Karatsuba post-processing
+	veor		$t2,$Xl,$Xh
+	veor		$Xm,$Xm,$t1
+	 vld1.64	{$t1},[$inp],$inc	@ load [rotated] inp
+	veor		$Xm,$Xm,$t2
+	vpmull.p64	$t2,$Xl,$t3		@ 1st phase
+
+	vmov		$Xh#lo,$Xm#hi		@ Xh|Xm - 256-bit result
+	vmov		$Xm#hi,$Xl#lo		@ Xm is rotated Xl
+#ifndef __ARMEB__
+	 vrev64.8	$t1,$t1
+#endif
+	veor		$Xl,$Xm,$t2
+	 vext.8		$IN,$t1,$t1,#8
+
+	vext.8		$t2,$Xl,$Xl,#8		@ 2nd phase
+	vpmull.p64	$Xl,$Xl,$t3
+	veor		$t2,$t2,$Xh
+	veor		$Xl,$Xl,$t2
+	b.hs		.Loop_v8
+
+#ifndef __ARMEB__
+	vrev64.8	$Xl,$Xl
+#endif
+	vext.8		$Xl,$Xl,$Xl,#8
+	vst1.64		{$Xl},[$Xi]		@ write out Xi
+
+	ret
+.size	gcm_ghash_v8,.-gcm_ghash_v8
+___
+}
+$code.=<<___;
+.asciz  "GHASH for ARMv8, CRYPTOGAMS by <appro\@openssl.org>"
+.align  2
+___
+
+if ($flavour =~ /64/) {			######## 64-bit code
+    sub unvmov {
+	my $arg=shift;
+
+	$arg =~ m/q([0-9]+)#(lo|hi),\s*q([0-9]+)#(lo|hi)/o &&
+	sprintf	"ins	v%d.d[%d],v%d.d[%d]",$1,($2 eq "lo")?0:1,$3,($4 eq "lo")?0:1;
+    }
+    foreach(split("\n",$code)) {
+	s/cclr\s+([wx])([^,]+),\s*([a-z]+)/csel	$1$2,$1zr,$1$2,$3/o	or
+	s/vmov\.i8/movi/o		or	# fix up legacy mnemonics
+	s/vmov\s+(.*)/unvmov($1)/geo	or
+	s/vext\.8/ext/o			or
+	s/vshr\.s/sshr\.s/o		or
+	s/vshr/ushr/o			or
+	s/^(\s+)v/$1/o			or	# strip off v prefix
+	s/\bbx\s+lr\b/ret/o;
+
+	s/\bq([0-9]+)\b/"v".($1<8?$1:$1+8).".16b"/geo;	# old->new registers
+	s/@\s/\/\//o;				# old->new style commentary
+
+	# fix up remainig legacy suffixes
+	s/\.[ui]?8(\s)/$1/o;
+	s/\.[uis]?32//o and s/\.16b/\.4s/go;
+	m/\.p64/o and s/\.16b/\.1q/o;		# 1st pmull argument
+	m/l\.p64/o and s/\.16b/\.1d/go;		# 2nd and 3rd pmull arguments
+	s/\.[uisp]?64//o and s/\.16b/\.2d/go;
+	s/\.[42]([sd])\[([0-3])\]/\.$1\[$2\]/o;
+
+	print $_,"\n";
+    }
+} else {				######## 32-bit code
+    sub unvdup32 {
+	my $arg=shift;
+
+	$arg =~ m/q([0-9]+),\s*q([0-9]+)\[([0-3])\]/o &&
+	sprintf	"vdup.32	q%d,d%d[%d]",$1,2*$2+($3>>1),$3&1;
+    }
+    sub unvpmullp64 {
+	my ($mnemonic,$arg)=@_;
+
+	if ($arg =~ m/q([0-9]+),\s*q([0-9]+),\s*q([0-9]+)/o) {
+	    my $word = 0xf2a00e00|(($1&7)<<13)|(($1&8)<<19)
+				 |(($2&7)<<17)|(($2&8)<<4)
+				 |(($3&7)<<1) |(($3&8)<<2);
+	    $word |= 0x00010001	 if ($mnemonic =~ "2");
+	    # since ARMv7 instructions are always encoded little-endian.
+	    # correct solution is to use .inst directive, but older
+	    # assemblers don't implement it:-(
+	    sprintf ".byte\t0x%02x,0x%02x,0x%02x,0x%02x\t@ %s %s",
+			$word&0xff,($word>>8)&0xff,
+			($word>>16)&0xff,($word>>24)&0xff,
+			$mnemonic,$arg;
+	}
+    }
+
+    foreach(split("\n",$code)) {
+	s/\b[wx]([0-9]+)\b/r$1/go;		# new->old registers
+	s/\bv([0-9])\.[12468]+[bsd]\b/q$1/go;	# new->old registers
+        s/\/\/\s?/@ /o;				# new->old style commentary
+
+	# fix up remainig new-style suffixes
+	s/\],#[0-9]+/]!/o;
+
+	s/cclr\s+([^,]+),\s*([a-z]+)/mov$2	$1,#0/o			or
+	s/vdup\.32\s+(.*)/unvdup32($1)/geo				or
+	s/v?(pmull2?)\.p64\s+(.*)/unvpmullp64($1,$2)/geo		or
+	s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo	or
+	s/^(\s+)b\./$1b/o						or
+	s/^(\s+)ret/$1bx\tlr/o;
+
+        print $_,"\n";
+    }
+}
+
+close STDOUT; # enforce flush