[thirdparty/openssl.git] / crypto / bn / asm / sparcv9-gf2m.pl

#! /usr/bin/env perl
# Copyright 2012-2020 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
# 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/.
# ====================================================================
#
# October 2012
#
# The module implements bn_GF2m_mul_2x2 polynomial multiplication used
# in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
# the time being... Except that it has two code paths: one suitable
# for all SPARCv9 processors and one for VIS3-capable ones. Former
# delivers ~25-45% more, more for longer keys, heaviest DH and DSA
# verify operations on venerable UltraSPARC II. On T4 VIS3 code is
# ~100-230% faster than gcc-generated code and ~35-90% faster than
# the pure SPARCv9 code path.

$output = pop and open STDOUT,">$output";

$locals=16*8;

$tab="%l0";

@T=("%g2","%g3");
@i=("%g4","%g5");

($a1,$a2,$a4,$a8,$a12,$a48)=map("%o$_",(0..5));
($lo,$hi,$b)=("%g1",$a8,"%o7"); $a=$lo;

$code.=<<___;
#ifndef __ASSEMBLER__
# define __ASSEMBLER__ 1
#endif
#include "crypto/sparc_arch.h"

#ifdef __arch64__
.register	%g2,#scratch
.register	%g3,#scratch
#endif

#ifdef __PIC__
SPARC_PIC_THUNK(%g1)
#endif

.globl	bn_GF2m_mul_2x2
.align	16
bn_GF2m_mul_2x2:
        SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
        ld	[%g1+0],%g1             	! OPENSSL_sparcv9cap_P[0]

        andcc	%g1, SPARCV9_VIS3, %g0
        bz,pn	%icc,.Lsoftware
        nop

	sllx	%o1, 32, %o1
	sllx	%o3, 32, %o3
	or	%o2, %o1, %o1
	or	%o4, %o3, %o3
	.word	0x95b262ab			! xmulx   %o1, %o3, %o2
	.word	0x99b262cb			! xmulxhi %o1, %o3, %o4
	srlx	%o2, 32, %o1			! 13 cycles later
	st	%o2, [%o0+0]
	st	%o1, [%o0+4]
	srlx	%o4, 32, %o3
	st	%o4, [%o0+8]
	retl
	st	%o3, [%o0+12]

.align	16
.Lsoftware:
	save	%sp,-STACK_FRAME-$locals,%sp

	sllx	%i1,32,$a
	mov	-1,$a12
	sllx	%i3,32,$b
	or	%i2,$a,$a
	srlx	$a12,1,$a48			! 0x7fff...
	or	%i4,$b,$b
	srlx	$a12,2,$a12			! 0x3fff...
	add	%sp,STACK_BIAS+STACK_FRAME,$tab

	sllx	$a,2,$a4
	mov	$a,$a1
	sllx	$a,1,$a2

	srax	$a4,63,@i[1]			! broadcast 61st bit
	and	$a48,$a4,$a4			! (a<<2)&0x7fff...
	srlx	$a48,2,$a48
	srax	$a2,63,@i[0]			! broadcast 62nd bit
	and	$a12,$a2,$a2			! (a<<1)&0x3fff...
	srax	$a1,63,$lo			! broadcast 63rd bit
	and	$a48,$a1,$a1			! (a<<0)&0x1fff...

	sllx	$a1,3,$a8
	and	$b,$lo,$lo
	and	$b,@i[0],@i[0]
	and	$b,@i[1],@i[1]

	stx	%g0,[$tab+0*8]			! tab[0]=0
	xor	$a1,$a2,$a12
	stx	$a1,[$tab+1*8]			! tab[1]=a1
	stx	$a2,[$tab+2*8]			! tab[2]=a2
	 xor	$a4,$a8,$a48
	stx	$a12,[$tab+3*8]			! tab[3]=a1^a2
	 xor	$a4,$a1,$a1

	stx	$a4,[$tab+4*8]			! tab[4]=a4
	xor	$a4,$a2,$a2
	stx	$a1,[$tab+5*8]			! tab[5]=a1^a4
	xor	$a4,$a12,$a12
	stx	$a2,[$tab+6*8]			! tab[6]=a2^a4
	 xor	$a48,$a1,$a1
	stx	$a12,[$tab+7*8]			! tab[7]=a1^a2^a4
	 xor	$a48,$a2,$a2

	stx	$a8,[$tab+8*8]			! tab[8]=a8
	xor	$a48,$a12,$a12
	stx	$a1,[$tab+9*8]			! tab[9]=a1^a8
	 xor	$a4,$a1,$a1
	stx	$a2,[$tab+10*8]			! tab[10]=a2^a8
	 xor	$a4,$a2,$a2
	stx	$a12,[$tab+11*8]		! tab[11]=a1^a2^a8

	xor	$a4,$a12,$a12
	stx	$a48,[$tab+12*8]		! tab[12]=a4^a8
	 srlx	$lo,1,$hi
	stx	$a1,[$tab+13*8]			! tab[13]=a1^a4^a8
	 sllx	$lo,63,$lo
	stx	$a2,[$tab+14*8]			! tab[14]=a2^a4^a8
	 srlx	@i[0],2,@T[0]
	stx	$a12,[$tab+15*8]		! tab[15]=a1^a2^a4^a8

	sllx	@i[0],62,$a1
	 sllx	$b,3,@i[0]
	srlx	@i[1],3,@T[1]
	 and	@i[0],`0xf<<3`,@i[0]
	sllx	@i[1],61,$a2
	 ldx	[$tab+@i[0]],@i[0]
	 srlx	$b,4-3,@i[1]
	xor	@T[0],$hi,$hi
	 and	@i[1],`0xf<<3`,@i[1]
	xor	$a1,$lo,$lo
	 ldx	[$tab+@i[1]],@i[1]
	xor	@T[1],$hi,$hi

	xor	@i[0],$lo,$lo
	srlx	$b,8-3,@i[0]
	 xor	$a2,$lo,$lo
	and	@i[0],`0xf<<3`,@i[0]
___
for($n=1;$n<14;$n++) {
$code.=<<___;
	sllx	@i[1],`$n*4`,@T[0]
	ldx	[$tab+@i[0]],@i[0]
	srlx	@i[1],`64-$n*4`,@T[1]
	xor	@T[0],$lo,$lo
	srlx	$b,`($n+2)*4`-3,@i[1]
	xor	@T[1],$hi,$hi
	and	@i[1],`0xf<<3`,@i[1]
___
	push(@i,shift(@i)); push(@T,shift(@T));
}
$code.=<<___;
	sllx	@i[1],`$n*4`,@T[0]
	ldx	[$tab+@i[0]],@i[0]
	srlx	@i[1],`64-$n*4`,@T[1]
	xor	@T[0],$lo,$lo

	sllx	@i[0],`($n+1)*4`,@T[0]
	 xor	@T[1],$hi,$hi
	srlx	@i[0],`64-($n+1)*4`,@T[1]
	xor	@T[0],$lo,$lo
	xor	@T[1],$hi,$hi

	srlx	$lo,32,%i1
	st	$lo,[%i0+0]
	st	%i1,[%i0+4]
	srlx	$hi,32,%i2
	st	$hi,[%i0+8]
	st	%i2,[%i0+12]

	ret
	restore
.type	bn_GF2m_mul_2x2,#function
.size	bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
.asciz	"GF(2^m) Multiplication for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
.align	4
___

$code =~ s/\`([^\`]*)\`/eval($1)/gem;
print $code;
close STDOUT or die "error closing STDOUT: $!";
Commit	Line	Data
6aa36e8e	1	#! /usr/bin/env perl
33388b44	2	# Copyright 2012-2020 The OpenSSL Project Authors. All Rights Reserved.
6aa36e8e	3	#
367ace68	4	# Licensed under the Apache License 2.0 (the "License"). You may not use
6aa36e8e RS	5	# this file except in compliance with the License. You can obtain a copy
	6	# in the file LICENSE in the source distribution or at
	7	# https://www.openssl.org/source/license.html
	8
0c832ec5 AP	9	#
	10	# ====================================================================
	11	# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
	12	# project. The module is, however, dual licensed under OpenSSL and
	13	# CRYPTOGAMS licenses depending on where you obtain it. For further
	14	# details see http://www.openssl.org/~appro/cryptogams/.
	15	# ====================================================================
	16	#
	17	# October 2012
	18	#
	19	# The module implements bn_GF2m_mul_2x2 polynomial multiplication used
	20	# in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
	21	# the time being... Except that it has two code paths: one suitable
	22	# for all SPARCv9 processors and one for VIS3-capable ones. Former
	23	# delivers ~25-45% more, more for longer keys, heaviest DH and DSA
	24	# verify operations on venerable UltraSPARC II. On T4 VIS3 code is
	25	# ~100-230% faster than gcc-generated code and ~35-90% faster than
	26	# the pure SPARCv9 code path.
	27
1aa89a7a	28	$output = pop and open STDOUT,">$output";
6bd7a4d9	29
0c832ec5 AP	30	$locals=16*8;
0c832ec5 AP	31
0c832ec5 AP	32	$tab="%l0";
	33
	34	@T=("%g2","%g3");
	35	@i=("%g4","%g5");
	36
	37	($a1,$a2,$a4,$a8,$a12,$a48)=map("%o$_",(0..5));
	38	($lo,$hi,$b)=("%g1",$a8,"%o7"); $a=$lo;
	39
	40	$code.=<<___;
52f7e44e TM	41	#ifndef __ASSEMBLER__
	42	# define __ASSEMBLER__ 1
	43	#endif
	44	#include "crypto/sparc_arch.h"
1efd5830 AP	45
	46	#ifdef __arch64__
	47	.register %g2,#scratch
	48	.register %g3,#scratch
	49	#endif
	50
0c832ec5 AP	51	#ifdef __PIC__
	52	SPARC_PIC_THUNK(%g1)
	53	#endif
	54
	55	.globl bn_GF2m_mul_2x2
	56	.align 16
	57	bn_GF2m_mul_2x2:
	58	SPARC_LOAD_ADDRESS_LEAF(OPENSSL_sparcv9cap_P,%g1,%g5)
	59	ld [%g1+0],%g1 ! OPENSSL_sparcv9cap_P[0]
	60
	61	andcc %g1, SPARCV9_VIS3, %g0
	62	bz,pn %icc,.Lsoftware
	63	nop
	64
	65	sllx %o1, 32, %o1
	66	sllx %o3, 32, %o3
	67	or %o2, %o1, %o1
	68	or %o4, %o3, %o3
	69	.word 0x95b262ab ! xmulx %o1, %o3, %o2
	70	.word 0x99b262cb ! xmulxhi %o1, %o3, %o4
	71	srlx %o2, 32, %o1 ! 13 cycles later
	72	st %o2, [%o0+0]
	73	st %o1, [%o0+4]
	74	srlx %o4, 32, %o3
	75	st %o4, [%o0+8]
	76	retl
	77	st %o3, [%o0+12]
	78
	79	.align 16
	80	.Lsoftware:
1efd5830	81	save %sp,-STACK_FRAME-$locals,%sp
0c832ec5 AP	82
	83	sllx %i1,32,$a
	84	mov -1,$a12
	85	sllx %i3,32,$b
	86	or %i2,$a,$a
	87	srlx $a12,1,$a48 ! 0x7fff...
	88	or %i4,$b,$b
	89	srlx $a12,2,$a12 ! 0x3fff...
1efd5830	90	add %sp,STACK_BIAS+STACK_FRAME,$tab
0c832ec5 AP	91
	92	sllx $a,2,$a4
	93	mov $a,$a1
	94	sllx $a,1,$a2
	95
	96	srax $a4,63,@i[1] ! broadcast 61st bit
	97	and $a48,$a4,$a4 ! (a<<2)&0x7fff...
	98	srlx $a48,2,$a48
	99	srax $a2,63,@i[0] ! broadcast 62nd bit
	100	and $a12,$a2,$a2 ! (a<<1)&0x3fff...
	101	srax $a1,63,$lo ! broadcast 63rd bit
	102	and $a48,$a1,$a1 ! (a<<0)&0x1fff...
	103
	104	sllx $a1,3,$a8
	105	and $b,$lo,$lo
	106	and $b,@i[0],@i[0]
	107	and $b,@i[1],@i[1]
	108
	109	stx %g0,[$tab+0*8] ! tab[0]=0
	110	xor $a1,$a2,$a12
	111	stx $a1,[$tab+1*8] ! tab[1]=a1
	112	stx $a2,[$tab+2*8] ! tab[2]=a2
	113	xor $a4,$a8,$a48
	114	stx $a12,[$tab+3*8] ! tab[3]=a1^a2
	115	xor $a4,$a1,$a1
	116
	117	stx $a4,[$tab+4*8] ! tab[4]=a4
	118	xor $a4,$a2,$a2
	119	stx $a1,[$tab+5*8] ! tab[5]=a1^a4
	120	xor $a4,$a12,$a12
	121	stx $a2,[$tab+6*8] ! tab[6]=a2^a4
	122	xor $a48,$a1,$a1
	123	stx $a12,[$tab+7*8] ! tab[7]=a1^a2^a4
	124	xor $a48,$a2,$a2
	125
	126	stx $a8,[$tab+8*8] ! tab[8]=a8
	127	xor $a48,$a12,$a12
	128	stx $a1,[$tab+9*8] ! tab[9]=a1^a8
	129	xor $a4,$a1,$a1
	130	stx $a2,[$tab+10*8] ! tab[10]=a2^a8
	131	xor $a4,$a2,$a2
	132	stx $a12,[$tab+11*8] ! tab[11]=a1^a2^a8
	133
	134	xor $a4,$a12,$a12
	135	stx $a48,[$tab+12*8] ! tab[12]=a4^a8
	136	srlx $lo,1,$hi
	137	stx $a1,[$tab+13*8] ! tab[13]=a1^a4^a8
	138	sllx $lo,63,$lo
	139	stx $a2,[$tab+14*8] ! tab[14]=a2^a4^a8
	140	srlx @i[0],2,@T[0]
	141	stx $a12,[$tab+15*8] ! tab[15]=a1^a2^a4^a8
	142
	143	sllx @i[0],62,$a1
	144	sllx $b,3,@i[0]
	145	srlx @i[1],3,@T[1]
	146	and @i[0],`0xf<<3`,@i[0]
	147	sllx @i[1],61,$a2
	148	ldx [$tab+@i[0]],@i[0]
	149	srlx $b,4-3,@i[1]
	150	xor @T[0],$hi,$hi
	151	and @i[1],`0xf<<3`,@i[1]
	152	xor $a1,$lo,$lo
	153	ldx [$tab+@i[1]],@i[1]
	154	xor @T[1],$hi,$hi
155
156	xor @i[0],$lo,$lo
157	srlx $b,8-3,@i[0]
158	xor $a2,$lo,$lo
159	and @i[0],`0xf<<3`,@i[0]
160	___
161	for($n=1;$n<14;$n++) {
162	$code.=<<___;
163	sllx @i[1],`$n*4`,@T[0]
164	ldx [$tab+@i[0]],@i[0]
165	srlx @i[1],`64-$n*4`,@T[1]
166	xor @T[0],$lo,$lo
167	srlx $b,`($n+2)*4`-3,@i[1]
168	xor @T[1],$hi,$hi
169	and @i[1],`0xf<<3`,@i[1]
170	___
171	push(@i,shift(@i)); push(@T,shift(@T));
172	}
173	$code.=<<___;
174	sllx @i[1],`$n*4`,@T[0]
175	ldx [$tab+@i[0]],@i[0]
176	srlx @i[1],`64-$n*4`,@T[1]
177	xor @T[0],$lo,$lo
178
179	sllx @i[0],`($n+1)*4`,@T[0]
180	xor @T[1],$hi,$hi
181	srlx @i[0],`64-($n+1)*4`,@T[1]
182	xor @T[0],$lo,$lo
183	xor @T[1],$hi,$hi
184
185	srlx $lo,32,%i1
186	st $lo,[%i0+0]
187	st %i1,[%i0+4]
188	srlx $hi,32,%i2
189	st $hi,[%i0+8]
190	st %i2,[%i0+12]
191
192	ret
193	restore
194	.type bn_GF2m_mul_2x2,#function
195	.size bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
196	.asciz "GF(2^m) Multiplication for SPARCv9, CRYPTOGAMS by <appro\@openssl.org>"
197	.align 4
198	___
199
200	$code =~ s/\`([^\`]*)\`/eval($1)/gem;
201	print $code;
a21314db	202	close STDOUT or die "error closing STDOUT: $!";