-#!/usr/bin/env perl
+#! /usr/bin/env perl
+# Copyright 2011-2020 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the Apache License 2.0 (the "License"). You may not use
+# this file except in compliance with the License. You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
#
# ====================================================================
# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
# length, more for longer keys. Even though NEON 1x1 multiplication
# runs in even less cycles, ~30, improvement is measurable only on
# longer keys. One has to optimize code elsewhere to get NEON glow...
+#
+# April 2014
+#
+# Double bn_GF2m_mul_2x2 performance by using algorithm from paper
+# referred below, which improves ECDH and ECDSA verify benchmarks
+# by 18-40%.
+#
+# 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
+
+# $output is the last argument if it looks like a file (it has an extension)
+# $flavour is the first argument if it doesn't look like a file
+$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
+$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
-while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
-open STDOUT,">$output";
+if ($flavour && $flavour ne "void") {
+ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+ ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
+ ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
+ die "can't locate arm-xlate.pl";
-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".($1/2):""; }
+ open STDOUT,"| \"$^X\" $xlate $flavour \"$output\""
+ or die "can't call $xlate: $1";
+} else {
+ $output and open STDOUT,">$output";
+}
$code=<<___;
#include "arm_arch.h"
-.text
+#if defined(__thumb2__)
+.syntax unified
+.thumb
+#else
.code 32
-
-#if __ARM_ARCH__>=7
-.fpu neon
-
-.type mul_1x1_neon,%function
-.align 5
-mul_1x1_neon:
- vshl.u64 `&Dlo("q1")`,d16,#8 @ q1-q3 are slided $a
- vmull.p8 `&Q("d0")`,d16,d17 @ a·bb
- vshl.u64 `&Dlo("q2")`,d16,#16
- vmull.p8 q1,`&Dlo("q1")`,d17 @ a<<8·bb
- vshl.u64 `&Dlo("q3")`,d16,#24
- vmull.p8 q2,`&Dlo("q2")`,d17 @ a<<16·bb
- vshr.u64 `&Dlo("q1")`,#8
- vmull.p8 q3,`&Dlo("q3")`,d17 @ a<<24·bb
- vshl.u64 `&Dhi("q1")`,#24
- veor d0,`&Dlo("q1")`
- vshr.u64 `&Dlo("q2")`,#16
- veor d0,`&Dhi("q1")`
- vshl.u64 `&Dhi("q2")`,#16
- veor d0,`&Dlo("q2")`
- vshr.u64 `&Dlo("q3")`,#24
- veor d0,`&Dhi("q2")`
- vshl.u64 `&Dhi("q3")`,#8
- veor d0,`&Dlo("q3")`
- veor d0,`&Dhi("q3")`
- bx lr
-.size mul_1x1_neon,.-mul_1x1_neon
#endif
+
+.text
___
################
# private interface to mul_1x1_ialu
eor $hi,$hi,$t0,lsr#8
ldr $t0,[sp,$i0] @ tab[b >> 30 ]
+#ifdef __thumb2__
+ itt ne
+#endif
eorne $lo,$lo,$b,lsl#30
eorne $hi,$hi,$b,lsr#2
tst $a,#1<<31
eor $lo,$lo,$t1,lsl#27
eor $hi,$hi,$t1,lsr#5
+#ifdef __thumb2__
+ itt ne
+#endif
eorne $lo,$lo,$b,lsl#31
eorne $hi,$hi,$b,lsr#1
eor $lo,$lo,$t0,lsl#30
################
# void bn_GF2m_mul_2x2(BN_ULONG *r,
# BN_ULONG a1,BN_ULONG a0,
-# BN_ULONG b1,BN_ULONG b0); # r[3..0]=a1a0·b1b0
-
-($A1,$B1,$A0,$B0,$A1B1,$A0B0)=map("d$_",(18..23));
-
+# BN_ULONG b1,BN_ULONG b0); # r[3..0]=a1a0·b1b0
+{
$code.=<<___;
.global bn_GF2m_mul_2x2
.type bn_GF2m_mul_2x2,%function
.align 5
bn_GF2m_mul_2x2:
-#if __ARM_ARCH__>=7
+#if __ARM_MAX_ARCH__>=7
+ stmdb sp!,{r10,lr}
ldr r12,.LOPENSSL_armcap
-.Lpic: ldr r12,[pc,r12]
- tst r12,#1
- beq .Lialu
-
- veor $A1,$A1
- vmov.32 $B1,r3,r3 @ two copies of b1
- vmov.32 ${A1}[0],r1 @ a1
-
- veor $A0,$A0
- vld1.32 ${B0}[],[sp,:32] @ two copies of b0
- vmov.32 ${A0}[0],r2 @ a0
- mov r12,lr
-
- vmov d16,$A1
- vmov d17,$B1
- bl mul_1x1_neon @ a1·b1
- vmov $A1B1,d0
-
- vmov d16,$A0
- vmov d17,$B0
- bl mul_1x1_neon @ a0·b0
- vmov $A0B0,d0
-
- veor d16,$A0,$A1
- veor d17,$B0,$B1
- veor $A0,$A0B0,$A1B1
- bl mul_1x1_neon @ (a0+a1)·(b0+b1)
-
- veor d0,$A0 @ (a0+a1)·(b0+b1)-a0·b0-a1·b1
- vshl.u64 d1,d0,#32
- vshr.u64 d0,d0,#32
- veor $A0B0,d1
- veor $A1B1,d0
- vst1.32 {${A0B0}[0]},[r0,:32]!
- vst1.32 {${A0B0}[1]},[r0,:32]!
- vst1.32 {${A1B1}[0]},[r0,:32]!
- vst1.32 {${A1B1}[1]},[r0,:32]
- bx r12
-.align 4
-.Lialu:
+# if !defined(_WIN32)
+ adr r10,.LOPENSSL_armcap
+ ldr r12,[r12,r10]
+# endif
+# if defined(__APPLE__) || defined(_WIN32)
+ ldr r12,[r12]
+# endif
+ tst r12,#ARMV7_NEON
+ itt ne
+ ldrne r10,[sp],#8
+ bne .LNEON
+ stmdb sp!,{r4-r9}
+#else
+ stmdb sp!,{r4-r10,lr}
#endif
___
$ret="r10"; # reassigned 1st argument
$code.=<<___;
- stmdb sp!,{r4-r10,lr}
mov $ret,r0 @ reassign 1st argument
mov $b,r3 @ $b=b1
+ sub r7,sp,#36
+ mov r8,sp
+ and r7,r7,#-32
ldr r3,[sp,#32] @ load b0
mov $mask,#7<<2
- sub sp,#32 @ allocate tab[8]
+ mov sp,r7 @ allocate tab[8]
+ str r8,[r7,#32]
- bl mul_1x1_ialu @ a1·b1
+ bl mul_1x1_ialu @ a1·b1
str $lo,[$ret,#8]
str $hi,[$ret,#12]
- eor $b,r3 @ flip b0 and b1
- eor $a,r2 @ flip a0 and a1
- eor r3,$b
- eor r2,$a
- eor $b,r3
- eor $a,r2
- bl mul_1x1_ialu @ a0·b0
+ eor $b,$b,r3 @ flip b0 and b1
+ eor $a,$a,r2 @ flip a0 and a1
+ eor r3,r3,$b
+ eor r2,r2,$a
+ eor $b,$b,r3
+ eor $a,$a,r2
+ bl mul_1x1_ialu @ a0·b0
str $lo,[$ret]
str $hi,[$ret,#4]
- eor $a,r2
- eor $b,r3
- bl mul_1x1_ialu @ (a1+a0)·(b1+b0)
+ eor $a,$a,r2
+ eor $b,$b,r3
+ bl mul_1x1_ialu @ (a1+a0)·(b1+b0)
___
@r=map("r$_",(6..9));
$code.=<<___;
ldmia $ret,{@r[0]-@r[3]}
- eor $lo,$hi
- eor $hi,@r[1]
- eor $lo,@r[0]
- eor $hi,@r[2]
- eor $lo,@r[3]
- eor $hi,@r[3]
+ eor $lo,$lo,$hi
+ ldr sp,[sp,#32] @ destroy tab[8]
+ eor $hi,$hi,@r[1]
+ eor $lo,$lo,@r[0]
+ eor $hi,$hi,@r[2]
+ eor $lo,$lo,@r[3]
+ eor $hi,$hi,@r[3]
str $hi,[$ret,#8]
- eor $lo,$hi
- add sp,#32 @ destroy tab[8]
+ eor $lo,$lo,$hi
str $lo,[$ret,#4]
#if __ARM_ARCH__>=5
moveq pc,lr @ be binary compatible with V4, yet
bx lr @ interoperable with Thumb ISA:-)
#endif
+___
+}
+{
+my ($r,$t0,$t1,$t2,$t3)=map("q$_",(0..3,8..12));
+my ($a,$b,$k48,$k32,$k16)=map("d$_",(26..31));
+
+$code.=<<___;
+#if __ARM_MAX_ARCH__>=7
+.arch armv7-a
+.fpu neon
+
+.align 5
+.LNEON:
+ ldr r12, [sp] @ 5th argument
+ vmov $a, r2, r1
+ vmov $b, r12, r3
+ vmov.i64 $k48, #0x0000ffffffffffff
+ vmov.i64 $k32, #0x00000000ffffffff
+ vmov.i64 $k16, #0x000000000000ffff
+
+ 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, $t2
+
+ vst1.32 {$r}, [r0]
+ ret @ bx lr
+#endif
+___
+}
+$code.=<<___;
.size bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
-#if __ARM_ARCH__>=7
+#if __ARM_MAX_ARCH__>=7
.align 5
.LOPENSSL_armcap:
-.word OPENSSL_armcap-(.Lpic+8)
+# ifdef _WIN32
+.word OPENSSL_armcap_P
+# else
+.word OPENSSL_armcap_P-.
+# endif
#endif
-.asciz "GF2m Multiplication for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
+.asciz "GF(2^m) Multiplication for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
.align 5
-.comm OPENSSL_armcap,4,4
+#if __ARM_MAX_ARCH__>=7
+.comm OPENSSL_armcap_P,4,4
+#endif
___
-$code =~ s/\`([^\`]*)\`/eval $1/gem;
-$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm; # make it possible to compile with -march=armv4
-print $code;
-close STDOUT; # enforce flush
+foreach (split("\n",$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 or die "error closing STDOUT: $!"; # enforce flush