invert_limb.asm

a very popular packet of cryptography tools,it encloses the most common used algorithm and protocols

dnl  Alpha mpn_invert_limb -- Invert a normalized limb.

dnl  Copyright 1996, 2000, 2001, 2002 Free Software Foundation, Inc.
dnl
dnl  This file is part of the GNU MP Library.
dnl
dnl  The GNU MP Library is free software; you can redistribute it and/or modify
dnl  it under the terms of the GNU Lesser General Public License as published
dnl  by the Free Software Foundation; either version 2.1 of the License, or (at
dnl  your option) any later version.
dnl
dnl  The GNU MP Library is distributed in the hope that it will be useful, but
dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
dnl  License for more details.
dnl
dnl  You should have received a copy of the GNU Lesser General Public License
dnl  along with the GNU MP Library; see the file COPYING.LIB.  If not, write to
dnl  the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
dnl  MA 02111-1307, USA.


dnl  This is based on ideas of Peter L. Montgomery.
dnl
dnl  The table below uses 4096 bytes.  The file mentioned above has an
dnl  alternative function that doesn't require the table, but it runs 50%
dnl  slower than this.

include(`../config.m4')

ASM_START()

FLOAT64($C36,9223372036854775808.0)		C 2^63

PROLOGUE(mpn_invert_limb,gp)
	lda	r30,-16(r30)
	addq	r16,r16,r1
	bne	r1,$73
	lda	r0,-1
	br	r31,$Lend
$73:
	srl	r16,1,r1
	stq	r1,0(r30)
	ldt	f11,0(r30)
	cvtqt	f11,f1
	LEA(r1,$C36)
	ldt	f10,0(r1)
	divt	f10,f1,f10
	LEA(r2,$invtab-4096)
	srl	r16,52,r1		C extract high 8 bits
	addq	r1,r1,r1		C align ...0000bbbbbbbb0
	addq	r1,r2,r1		C compute array offset
	ldq_u	r2,0(r1)		C load quadword containing our 16 bits
bigend(`addq	r1,1,r1')
	extwl	r2,r1,r2		C extract desired 16 bits
	sll	r2,48,r0
	umulh	r16,r0,r1
	addq	r16,r1,r3
	stq	r3,0(r30)
	ldt	f11,0(r30)
	cvtqt	f11,f1
	mult	f1,f10,f1
	cvttqc	f1,f1
	stt	f1,0(r30)
	ldq	r4,0(r30)
	subq	r0,r4,r0
	umulh	r16,r0,r1
	mulq	r16,r0,r2
	addq	r16,r1,r3
	bge	r3,$Loop2
$Loop1:	addq	r2,r16,r2
	cmpult	r2,r16,r1
	addq	r3,r1,r3
	addq	r0,1,r0
	blt	r3,$Loop1
$Loop2:	cmpult	r2,r16,r1
	subq	r0,1,r0
	subq	r3,r1,r3
	subq	r2,r16,r2
	bge	r3,$Loop2
$Lend:
	lda	r30,16(r30)
	ret	r31,(r26),1
EPILOGUE(mpn_invert_limb)
DATASTART($invtab)
	.word 0xffff,0xffc0,0xff80,0xff40,0xff00,0xfec0,0xfe81,0xfe41
	.word 0xfe01,0xfdc2,0xfd83,0xfd43,0xfd04,0xfcc5,0xfc86,0xfc46
	.word 0xfc07,0xfbc8,0xfb8a,0xfb4b,0xfb0c,0xfacd,0xfa8e,0xfa50
	.word 0xfa11,0xf9d3,0xf994,0xf956,0xf918,0xf8d9,0xf89b,0xf85d
	.word 0xf81f,0xf7e1,0xf7a3,0xf765,0xf727,0xf6ea,0xf6ac,0xf66e
	.word 0xf631,0xf5f3,0xf5b6,0xf578,0xf53b,0xf4fd,0xf4c0,0xf483
	.word 0xf446,0xf409,0xf3cc,0xf38f,0xf352,0xf315,0xf2d8,0xf29c
	.word 0xf25f,0xf222,0xf1e6,0xf1a9,0xf16d,0xf130,0xf0f4,0xf0b8
	.word 0xf07c,0xf03f,0xf003,0xefc7,0xef8b,0xef4f,0xef14,0xeed8
	.word 0xee9c,0xee60,0xee25,0xede9,0xedae,0xed72,0xed37,0xecfb
	.word 0xecc0,0xec85,0xec4a,0xec0e,0xebd3,0xeb98,0xeb5d,0xeb22
	.word 0xeae8,0xeaad,0xea72,0xea37,0xe9fd,0xe9c2,0xe988,0xe94d
	.word 0xe913,0xe8d8,0xe89e,0xe864,0xe829,0xe7ef,0xe7b5,0xe77b
	.word 0xe741,0xe707,0xe6cd,0xe694,0xe65a,0xe620,0xe5e6,0xe5ad
	.word 0xe573,0xe53a,0xe500,0xe4c7,0xe48d,0xe454,0xe41b,0xe3e2
	.word 0xe3a9,0xe370,0xe336,0xe2fd,0xe2c5,0xe28c,0xe253,0xe21a
	.word 0xe1e1,0xe1a9,0xe170,0xe138,0xe0ff,0xe0c7,0xe08e,0xe056
	.word 0xe01e,0xdfe5,0xdfad,0xdf75,0xdf3d,0xdf05,0xdecd,0xde95
	.word 0xde5d,0xde25,0xdded,0xddb6,0xdd7e,0xdd46,0xdd0f,0xdcd7
	.word 0xdca0,0xdc68,0xdc31,0xdbf9,0xdbc2,0xdb8b,0xdb54,0xdb1d
	.word 0xdae6,0xdaae,0xda78,0xda41,0xda0a,0xd9d3,0xd99c,0xd965
	.word 0xd92f,0xd8f8,0xd8c1,0xd88b,0xd854,0xd81e,0xd7e8,0xd7b1
	.word 0xd77b,0xd745,0xd70e,0xd6d8,0xd6a2,0xd66c,0xd636,0xd600
	.word 0xd5ca,0xd594,0xd55f,0xd529,0xd4f3,0xd4bd,0xd488,0xd452
	.word 0xd41d,0xd3e7,0xd3b2,0xd37c,0xd347,0xd312,0xd2dd,0xd2a7
	.word 0xd272,0xd23d,0xd208,0xd1d3,0xd19e,0xd169,0xd134,0xd100
	.word 0xd0cb,0xd096,0xd061,0xd02d,0xcff8,0xcfc4,0xcf8f,0xcf5b
	.word 0xcf26,0xcef2,0xcebe,0xce89,0xce55,0xce21,0xcded,0xcdb9
	.word 0xcd85,0xcd51,0xcd1d,0xcce9,0xccb5,0xcc81,0xcc4e,0xcc1a
	.word 0xcbe6,0xcbb3,0xcb7f,0xcb4c,0xcb18,0xcae5,0xcab1,0xca7e
	.word 0xca4b,0xca17,0xc9e4,0xc9b1,0xc97e,0xc94b,0xc918,0xc8e5
	.word 0xc8b2,0xc87f,0xc84c,0xc819,0xc7e7,0xc7b4,0xc781,0xc74f
	.word 0xc71c,0xc6e9,0xc6b7,0xc684,0xc652,0xc620,0xc5ed,0xc5bb
	.word 0xc589,0xc557,0xc524,0xc4f2,0xc4c0,0xc48e,0xc45c,0xc42a
	.word 0xc3f8,0xc3c7,0xc395,0xc363,0xc331,0xc300,0xc2ce,0xc29c
	.word 0xc26b,0xc239,0xc208,0xc1d6,0xc1a5,0xc174,0xc142,0xc111
	.word 0xc0e0,0xc0af,0xc07e,0xc04d,0xc01c,0xbfeb,0xbfba,0xbf89
	.word 0xbf58,0xbf27,0xbef6,0xbec5,0xbe95,0xbe64,0xbe33,0xbe03
	.word 0xbdd2,0xbda2,0xbd71,0xbd41,0xbd10,0xbce0,0xbcb0,0xbc80
	.word 0xbc4f,0xbc1f,0xbbef,0xbbbf,0xbb8f,0xbb5f,0xbb2f,0xbaff
	.word 0xbacf,0xba9f,0xba6f,0xba40,0xba10,0xb9e0,0xb9b1,0xb981
	.word 0xb951,0xb922,0xb8f2,0xb8c3,0xb894,0xb864,0xb835,0xb806
	.word 0xb7d6,0xb7a7,0xb778,0xb749,0xb71a,0xb6eb,0xb6bc,0xb68d
	.word 0xb65e,0xb62f,0xb600,0xb5d1,0xb5a2,0xb574,0xb545,0xb516
	.word 0xb4e8,0xb4b9,0xb48a,0xb45c,0xb42e,0xb3ff,0xb3d1,0xb3a2
	.word 0xb374,0xb346,0xb318,0xb2e9,0xb2bb,0xb28d,0xb25f,0xb231
	.word 0xb203,0xb1d5,0xb1a7,0xb179,0xb14b,0xb11d,0xb0f0,0xb0c2
	.word 0xb094,0xb067,0xb039,0xb00b,0xafde,0xafb0,0xaf83,0xaf55
	.word 0xaf28,0xaefb,0xaecd,0xaea0,0xae73,0xae45,0xae18,0xadeb
	.word 0xadbe,0xad91,0xad64,0xad37,0xad0a,0xacdd,0xacb0,0xac83
	.word 0xac57,0xac2a,0xabfd,0xabd0,0xaba4,0xab77,0xab4a,0xab1e
	.word 0xaaf1,0xaac5,0xaa98,0xaa6c,0xaa40,0xaa13,0xa9e7,0xa9bb
	.word 0xa98e,0xa962,0xa936,0xa90a,0xa8de,0xa8b2,0xa886,0xa85a
	.word 0xa82e,0xa802,0xa7d6,0xa7aa,0xa77e,0xa753,0xa727,0xa6fb
	.word 0xa6d0,0xa6a4,0xa678,0xa64d,0xa621,0xa5f6,0xa5ca,0xa59f
	.word 0xa574,0xa548,0xa51d,0xa4f2,0xa4c6,0xa49b,0xa470,0xa445
	.word 0xa41a,0xa3ef,0xa3c4,0xa399,0xa36e,0xa343,0xa318,0xa2ed
	.word 0xa2c2,0xa297,0xa26d,0xa242,0xa217,0xa1ed,0xa1c2,0xa197
	.word 0xa16d,0xa142,0xa118,0xa0ed,0xa0c3,0xa098,0xa06e,0xa044
	.word 0xa01a,0x9fef,0x9fc5,0x9f9b,0x9f71,0x9f47,0x9f1c,0x9ef2
	.word 0x9ec8,0x9e9e,0x9e74,0x9e4b,0x9e21,0x9df7,0x9dcd,0x9da3
	.word 0x9d79,0x9d50,0x9d26,0x9cfc,0x9cd3,0x9ca9,0x9c80,0x9c56
	.word 0x9c2d,0x9c03,0x9bda,0x9bb0,0x9b87,0x9b5e,0x9b34,0x9b0b
	.word 0x9ae2,0x9ab9,0x9a8f,0x9a66,0x9a3d,0x9a14,0x99eb,0x99c2
	.word 0x9999,0x9970,0x9947,0x991e,0x98f6,0x98cd,0x98a4,0x987b
	.word 0x9852,0x982a,0x9801,0x97d8,0x97b0,0x9787,0x975f,0x9736
	.word 0x970e,0x96e5,0x96bd,0x9695,0x966c,0x9644,0x961c,0x95f3
	.word 0x95cb,0x95a3,0x957b,0x9553,0x952b,0x9503,0x94db,0x94b3
	.word 0x948b,0x9463,0x943b,0x9413,0x93eb,0x93c3,0x939b,0x9374
	.word 0x934c,0x9324,0x92fd,0x92d5,0x92ad,0x9286,0x925e,0x9237
	.word 0x920f,0x91e8,0x91c0,0x9199,0x9172,0x914a,0x9123,0x90fc
	.word 0x90d4,0x90ad,0x9086,0x905f,0x9038,0x9011,0x8fea,0x8fc3
	.word 0x8f9c,0x8f75,0x8f4e,0x8f27,0x8f00,0x8ed9,0x8eb2,0x8e8b
	.word 0x8e65,0x8e3e,0x8e17,0x8df1,0x8dca,0x8da3,0x8d7d,0x8d56
	.word 0x8d30,0x8d09,0x8ce3,0x8cbc,0x8c96,0x8c6f,0x8c49,0x8c23
	.word 0x8bfc,0x8bd6,0x8bb0,0x8b8a,0x8b64,0x8b3d,0x8b17,0x8af1
	.word 0x8acb,0x8aa5,0x8a7f,0x8a59,0x8a33,0x8a0d,0x89e7,0x89c1
	.word 0x899c,0x8976,0x8950,0x892a,0x8904,0x88df,0x88b9,0x8893
	.word 0x886e,0x8848,0x8823,0x87fd,0x87d8,0x87b2,0x878d,0x8767
	.word 0x8742,0x871d,0x86f7,0x86d2,0x86ad,0x8687,0x8662,0x863d
	.word 0x8618,0x85f3,0x85ce,0x85a9,0x8583,0x855e,0x8539,0x8514
	.word 0x84f0,0x84cb,0x84a6,0x8481,0x845c,0x8437,0x8412,0x83ee
	.word 0x83c9,0x83a4,0x8380,0x835b,0x8336,0x8312,0x82ed,0x82c9
	.word 0x82a4,0x8280,0x825b,0x8237,0x8212,0x81ee,0x81ca,0x81a5
	.word 0x8181,0x815d,0x8138,0x8114,0x80f0,0x80cc,0x80a8,0x8084
	.word 0x8060,0x803c,0x8018,0x7ff4,0x7fd0,0x7fac,0x7f88,0x7f64
	.word 0x7f40,0x7f1c,0x7ef8,0x7ed4,0x7eb1,0x7e8d,0x7e69,0x7e45