rockbox/utils/zenutils/libraries/beecrypt-4.1.2/beecrypt/mp.h
Nicolas Pennequin ca5bb76d2b Delete the svn:executable property and set svn:eol-style to native for all those text files.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@18012 a1c6a512-1295-4272-9138-f99709370657
2008-07-11 16:51:25 +00:00

691 lines
24 KiB
C

/*
* Copyright (c) 2002, 2003 Bob Deblier
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
*/
/*!\file mp.h
* \brief Multi-precision integer routines.
*
* The routines declared here are all low-level operations, most of them
* suitable to be implemented in assembler. Prime candidates are in order
* of importance (according to gprof):
* <ul>
* <li>mpaddmul
* <li>mpsetmul
* <li>mpaddsqrtrc
* <li>mpsub
* <li>mpadd
* </ul>
*
* With some smart use of available assembler instructions, it's possible
* to speed these routines up by a factor of 2 to 4.
*
* \author Bob Deblier <bob.deblier@pandora.be>
* \ingroup MP_m
*/
#ifndef _MP_H
#define _MP_H
#include "beecrypt/api.h"
#include "beecrypt/mpopt.h"
#define MP_HWBITS (MP_WBITS >> 1)
#define MP_WBYTES (MP_WBITS >> 3)
#define MP_WNIBBLES (MP_WBITS >> 2)
#if (MP_WBITS == 64)
# define MP_WORDS_TO_BITS(x) ((x) << 6)
# define MP_WORDS_TO_NIBBLES(x) ((x) << 4)
# define MP_WORDS_TO_BYTES(x) ((x) << 3)
# define MP_BITS_TO_WORDS(x) ((x) >> 6)
# define MP_NIBBLES_TO_WORDS(x) ((x) >> 4)
# define MP_BYTES_TO_WORDS(x) ((x) >> 3)
#elif (MP_WBITS == 32)
# define MP_WORDS_TO_BITS(x) ((x) << 5)
# define MP_WORDS_TO_NIBBLES(x) ((x) << 3)
# define MP_WORDS_TO_BYTES(x) ((x) << 2)
# define MP_BITS_TO_WORDS(x) ((x) >> 5)
# define MP_NIBBLES_TO_WORDS(x) ((x) >> 3)
# define MP_BYTES_TO_WORDS(x) ((x) >> 2)
#else
# error
#endif
#define MP_MSBMASK (((mpw) 0x1) << (MP_WBITS-1))
#define MP_LSBMASK ((mpw) 0x1)
#define MP_ALLMASK ~((mpw) 0x0)
#ifdef __cplusplus
extern "C" {
#endif
#ifndef ASM_MPCOPY
# define mpcopy(size, dst, src) memcpy(dst, src, MP_WORDS_TO_BYTES(size))
#else
BEECRYPTAPI
void mpcopy(size_t size, mpw* dest, const mpw* src);
#endif
#ifndef ASM_MPMOVE
# define mpmove(size, dst, src) memmove(dst, src, MP_WORDS_TO_BYTES(size))
#else
BEECRYPTAPI
void mpmove(size_t size, mpw* dest, const mpw* src);
#endif
/*!\fn void mpzero(size_t size, mpw* data)
* \brief This function zeroes a multi-precision integer of a given size.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
*/
BEECRYPTAPI
void mpzero(size_t size, mpw* data);
/*!\fn void mpfill(size_t size, mpw* data, mpw fill)
* \brief This function fills each word of a multi-precision integer with a
* given value.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \param fill The value fill the data with.
*/
BEECRYPTAPI
void mpfill(size_t size, mpw* data, mpw fill);
/*!\fn int mpodd(size_t size, const mpw* data)
* \brief This functions tests if a multi-precision integer is odd.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \retval 1 if odd
* \retval 0 if even
*/
BEECRYPTAPI
int mpodd (size_t size, const mpw* data);
/*!\fn int mpeven(size_t size, const mpw* data)
* \brief This function tests if a multi-precision integer is even.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \retval 1 if even
* \retval 0 if odd
*/
BEECRYPTAPI
int mpeven(size_t size, const mpw* data);
/*!\fn int mpz(size_t size, const mpw* data)
* \brief This function tests if a multi-precision integer is zero.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \retval 1 if zero
* \retval 0 if not zero
*/
BEECRYPTAPI
int mpz (size_t size, const mpw* data);
/*!\fn int mpnz(size_t size, const mpw* data)
* \brief This function tests if a multi-precision integer is not zero.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \retval 1 if not zero
* \retval 0 if zero
*/
BEECRYPTAPI
int mpnz (size_t size, const mpw* data);
/*!\fn int mpeq(size_t size, const mpw* xdata, const mpw* ydata)
* \brief This function tests if two multi-precision integers of the same size
* are equal.
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if equal
* \retval 0 if not equal
*/
BEECRYPTAPI
int mpeq (size_t size, const mpw* xdata, const mpw* ydata);
/*!\fn int mpne(size_t size, const mpw* xdata, const mpw* ydata)
* \brief This function tests if two multi-precision integers of the same size
* differ.
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if not equal
* \retval 0 if equal
*/
BEECRYPTAPI
int mpne (size_t size, const mpw* xdata, const mpw* ydata);
/*!\fn int mpgt(size_t size, const mpw* xdata, const mpw* ydata)
* \brief This function tests if the first of two multi-precision integers
* of the same size is greater than the second.
* \note The comparison treats the arguments as unsigned.
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if greater
* \retval 0 if less or equal
*/
BEECRYPTAPI
int mpgt (size_t size, const mpw* xdata, const mpw* ydata);
/*!\fn int mplt(size_t size, const mpw* xdata, const mpw* ydata)
* \brief This function tests if the first of two multi-precision integers
* of the same size is less than the second.
* \note The comparison treats the arguments as unsigned.
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if less
* \retval 0 if greater or equal
*/
BEECRYPTAPI
int mplt (size_t size, const mpw* xdata, const mpw* ydata);
/*!\fn int mpge(size_t size, const mpw* xdata, const mpw* ydata)
* \brief This function tests if the first of two multi-precision integers
* of the same size is greater than or equal to the second.
* \note The comparison treats the arguments as unsigned.
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if greater or equal
* \retval 0 if less
*/
BEECRYPTAPI
int mpge (size_t size, const mpw* xdata, const mpw* ydata);
/*!\fn int mple(size_t size, const mpw* xdata, const mpw* ydata)
* \brief This function tests if the first of two multi-precision integers
* of the same size is less than or equal to the second.
* \note The comparison treats the arguments as unsigned.
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if less or equal
* \retval 0 if greater
*/
BEECRYPTAPI
int mple (size_t size, const mpw* xdata, const mpw* ydata);
/*!\fn int mpeqx(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function tests if two multi-precision integers of different
* size are equal.
* \param xsize The size of the first multi-precision integer.
* \param xdata The first multi-precision integer.
* \param ysize The size of the first multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if equal
* \retval 0 if not equal
*/
BEECRYPTAPI
int mpeqx(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata);
/*!\fn int mpnex(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function tests if two multi-precision integers of different
* size are equal.
* \param xsize The size of the first multi-precision integer.
* \param xdata The first multi-precision integer.
* \param ysize The size of the first multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if equal
* \retval 0 if not equal
*/
BEECRYPTAPI
int mpnex(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata);
/*!\fn int mpgtx(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function tests if the first of two multi-precision integers
* of different size is greater than the second.
* \note The comparison treats the arguments as unsigned.
* \param xsize The size of the first multi-precision integer.
* \param xdata The first multi-precision integer.
* \param ysize The size of the second multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if greater
* \retval 0 if less or equal
*/
BEECRYPTAPI
int mpgtx(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata);
/*!\fn int mpltx(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function tests if the first of two multi-precision integers
* of different size is less than the second.
* \note The comparison treats the arguments as unsigned.
* \param xsize The size of the first multi-precision integer.
* \param xdata The first multi-precision integer.
* \param ysize The size of the second multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if less
* \retval 0 if greater or equal
*/
BEECRYPTAPI
int mpltx(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata);
/*!\fn int mpgex(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function tests if the first of two multi-precision integers
* of different size is greater than or equal to the second.
* \note The comparison treats the arguments as unsigned.
* \param xsize The size of the first multi-precision integer.
* \param xdata The first multi-precision integer.
* \param ysize The size of the second multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if greater or equal
* \retval 0 if less
*/
BEECRYPTAPI
int mpgex(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata);
/*!\fn int mplex(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function tests if the first of two multi-precision integers
* of different size is less than or equal to the second.
* \note The comparison treats the arguments as unsigned.
* \param xsize The size of the first multi-precision integer.
* \param xdata The first multi-precision integer.
* \param ysize The size of the second multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if less or equal
* \retval 0 if greater
*/
BEECRYPTAPI
int mplex(size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata);
/*!\fn int mpisone(size_t size, const mpw* data)
* \brief This functions tests if the value of a multi-precision integer is
* equal to one.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \retval 1 if one
* \retval 0 if not one
*/
BEECRYPTAPI
int mpisone(size_t size, const mpw* data);
/*!\fn int mpistwo(size_t size, const mpw* data)
* \brief This function tests if the value of a multi-precision integer is
* equal to two.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \retval 1 if two
* \retval 0 if not two
*/
BEECRYPTAPI
int mpistwo(size_t size, const mpw* data);
/*!\fn int mpleone(size_t size, const mpw* data);
* \brief This function tests if the value of a multi-precision integer is
* less than or equal to one.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \retval 1 if less than or equal to one.
* \retval 0 if greater than one.
*/
BEECRYPTAPI
int mpleone(size_t size, const mpw* data);
/*!\fn int mpeqmone(size_t size, const mpw* xdata, const mpw* ydata);
* \brief This function tests if multi-precision integer x is equal to y
* minus one.
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param ydata The second multi-precision integer.
* \retval 1 if less than or equal to one.
* \retval 0 if greater than one.
*/
BEECRYPTAPI
int mpeqmone(size_t size, const mpw* xdata, const mpw* ydata);
/*!\fn int mpmsbset(size_t size, const mpw* data)
* \brief This function tests if the most significant bit of a multi-precision
* integer is set.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \retval 1 if set
* \retval 0 if not set
*/
BEECRYPTAPI
int mpmsbset(size_t size, const mpw* data);
/*!\fn int mplsbset(size_t size, const mpw* data)
* \brief This function tests if the leiast significant bit of a multi-precision
* integer is set.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
* \retval 1 if set
* \retval 0 if not set
*/
BEECRYPTAPI
int mplsbset(size_t size, const mpw* data);
/*!\fn void mpsetmsb(size_t size, mpw* data)
* \brief This function sets the most significant bit of a multi-precision
* integer.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
*/
BEECRYPTAPI
void mpsetmsb(size_t size, mpw* data);
/*!\fn void mpsetlsb(size_t size, mpw* data)
* \brief This function sets the least significant bit of a multi-precision
* integer.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
*/
BEECRYPTAPI
void mpsetlsb(size_t size, mpw* data);
/*!\fn void mpclrmsb(size_t size, mpw* data)
* \brief This function clears the most significant bit of a multi-precision
* integer.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
*/
BEECRYPTAPI
void mpclrmsb(size_t size, mpw* data);
/*!\fn void mpclrlsb(size_t size, mpw* data)
* \brief This function clears the least significant bit of a multi-precision
* integer.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
*/
BEECRYPTAPI
void mpclrlsb(size_t size, mpw* data);
/*!\fn mpand(size_t size, mpw* xdata, const mpw* ydata)
* \brief This function computes the bit-wise AND of two multi-precision
* integers. Modifies xdata.
* \param size The size of the multi-precision integers.
* \param xdata The multi-precision integer data.
* \param ydata The multi-precision integer data.
*/
BEECRYPTAPI
void mpand(size_t size, mpw* xdata, const mpw* ydata);
/*!\fn void mpor(size_t size, mpw* xdata, const mpw* ydata)
* \brief This function computes the bit-wise OR of two multi-precision
* integers. Modifies xdata.
* \param size The size of the multi-precision integer.
* \param xdata The multi-precision integer data.
* \param ydata The multi-precision integer data.
*/
BEECRYPTAPI
void mpor(size_t size, mpw* xdata, const mpw* ydata);
/*!\fn void mpxor(size_t size, mpw* xdata, const mpw* ydata)
* \brief This function computes the bit-wise XOR of two multi-precision
* integers. Modifies xdata.
* \param size The size of the multi-precision integer.
* \param xdata The multi-precision integer data.
* \param ydata The multi-precision integer data.
*/
BEECRYPTAPI
void mpxor(size_t size, mpw* xdata, const mpw* ydata);
/*!\fn mpnot(size_t size, mpw* data)
* \brief This function flips all bits of a multi-precision integer.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
*/
BEECRYPTAPI
void mpnot(size_t size, mpw* data);
/*!\fn void mpsetw(size_t size, mpw* xdata, mpw y)
* \brief This function sets the value of a multi-precision integer to the
* given word. The given value is copied into the least significant word,
* while the most significant words are zeroed.
* \param size The size of the multi-precision integer.
* \param xdata The first multi-precision integer.
* \param y The multi-precision word.
*/
BEECRYPTAPI
void mpsetw(size_t size, mpw* xdata, mpw y);
/*!\fn void mpsetx(size_t xsize, mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function set the value of the first multi-precision integer
* to the second, truncating the most significant words if ysize > xsize, or
* zeroing the most significant words if ysize < xsize.
* \param xsize The size of the first multi-precision integer.
* \param xdata The first multi-precision integer.
* \param ysize The size of the second multi-precision integer.
* \param ydata The second multi-precision integer.
*/
void mpsetx(size_t xsize, mpw* xdata, size_t ysize, const mpw* ydata);
/*!\fn int mpaddw(size_t size, mpw* xdata, mpw y)
* \brief This function adds one word to a multi-precision integer.
* The performed operation is in pseudocode: x += y.
* \param size The size of the multi-precision integer.
* \param xdata The first multi-precision integer.
* \param y The multi-precision word.
* \return The carry-over value of the operation; this value is either 0 or 1.
*/
BEECRYPTAPI
int mpaddw(size_t size, mpw* xdata, mpw y);
/*!\fn int mpadd(size_t size, mpw* xdata, const mpw* ydata)
* \brief This function adds two multi-precision integers of equal size.
* The performed operation is in pseudocode: x += y.
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param ydata The second multi-precision integer.
* \return The carry-over value of the operation; this value is either 0 or 1.
*/
BEECRYPTAPI
int mpadd (size_t size, mpw* xdata, const mpw* ydata);
/*!\fn int mpaddx(size_t xsize, mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function adds two multi-precision integers of different size.
* The performed operation in pseudocode: x += y.
* \param xsize The size of the first multi-precision integer.
* \param xdata The first multi-precision integer.
* \param ysize The size of the second multi-precision integer.
* \param ydata The second multi-precision integer.
* \return The carry-over value of the operation; this value is either 0 or 1.
*/
BEECRYPTAPI
int mpaddx(size_t xsize, mpw* xdata, size_t ysize, const mpw* ydata);
/*!\fn int mpsubw(size_t size, mpw* xdata, mpw y)
* \brief This function subtracts one word to a multi-precision integer.
* The performed operation in pseudocode: x -= y
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param y The multi-precision word.
* \return The carry-over value of the operation; this value is either 0 or 1.
*/
BEECRYPTAPI
int mpsubw(size_t size, mpw* xdata, mpw y);
/*!\fn int mpsub(size_t size, mpw* xdata, const mpw* ydata)
* \brief This function subtracts two multi-precision integers of equal size.
* The performed operation in pseudocode: x -= y
* \param size The size of the multi-precision integers.
* \param xdata The first multi-precision integer.
* \param ydata The second multi-precision integer.
* \return The carry-over value of the operation; this value is either 0 or 1.
*/
BEECRYPTAPI
int mpsub (size_t size, mpw* xdata, const mpw* ydata);
/*!\fn int mpsubx(size_t xsize, mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function subtracts two multi-precision integers of different
* size. The performed operation in pseudocode: x -= y.
* \param xsize The size of the first multi-precision integer.
* \param xdata The first multi-precision integer.
* \param ysize The size of the second multi-precision integer.
* \param ydata The second multi-precision integer.
* \return The carry-over value of the operation; this value is either 0 or 1.
*/
BEECRYPTAPI
int mpsubx(size_t xsize, mpw* xdata, size_t ysize, const mpw* ydata);
BEECRYPTAPI
int mpmultwo(size_t size, mpw* data);
/*!\fn void mpneg(size_t size, mpw* data)
* \brief This function negates a multi-precision integer.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
*/
BEECRYPTAPI
void mpneg(size_t size, mpw* data);
/*!\fn size_t mpsize(size_t size, const mpw* data)
* \brief This function returns the true size of a multi-precision
* integer, after stripping leading zero words.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
*/
BEECRYPTAPI
size_t mpsize(size_t size, const mpw* data);
/*!\fn size_t mpbits(size_t size, const mpw* data)
* \brief This function returns the number of significant bits
* in a multi-precision integer.
* \param size The size of the multi-precision integer.
* \param data The multi-precision integer data.
*/
BEECRYPTAPI
size_t mpbits(size_t size, const mpw* data);
BEECRYPTAPI
size_t mpmszcnt(size_t size, const mpw* data);
BEECRYPTAPI
size_t mplszcnt(size_t size, const mpw* data);
BEECRYPTAPI
void mplshift(size_t size, mpw* data, size_t count);
BEECRYPTAPI
void mprshift(size_t size, mpw* data, size_t count);
BEECRYPTAPI
size_t mprshiftlsz(size_t size, mpw* data);
BEECRYPTAPI
size_t mpnorm(size_t size, mpw* data);
BEECRYPTAPI
void mpdivtwo (size_t size, mpw* data);
BEECRYPTAPI
void mpsdivtwo(size_t size, mpw* data);
/*!\fn mpw mpsetmul(size_t size, mpw* result, const mpw* data, mpw y)
* \brief This function performs a multi-precision multiply-setup.
*
* This function is used in the computation of a full multi-precision
* multiplication. By using it we can shave off a few cycles; otherwise we'd
* have to zero the least significant half of the result first and use
* another call to the slightly slower mpaddmul function.
*
* \param size The size of multi-precision integer multiplier.
* \param result The place where result will be accumulated.
* \param data The multi-precision integer multiplier.
* \param y The multiplicand.
* \return The carry-over multi-precision word.
*/
BEECRYPTAPI
mpw mpsetmul (size_t size, mpw* result, const mpw* data, mpw y);
/*!\fn mpw mpaddmul(size_t size, mpw* result, const mpw* data, mpw y)
* \brief This function performs a mult-precision multiply-accumulate.
*
* This function is used in the computation of a full multi-precision
* multiplication. It computes the product-by-one-word and accumulates it with
* the previous result.
*
* \param size The size of multi-precision integer multiplier.
* \param result The place where result will be accumulated.
* \param data The multi-precision integer multiplier.
* \param y The multiplicand.
* \retval The carry-over multi-precision word.
*/
BEECRYPTAPI
mpw mpaddmul (size_t size, mpw* result, const mpw* data, mpw y);
/*!\fn void mpaddsqrtrc(size_t size, mpw* result, const mpw* data)
* \brief This function is used in the calculation of a multi-precision
* squaring.
*/
BEECRYPTAPI
void mpaddsqrtrc(size_t size, mpw* result, const mpw* data);
/*!\fn void mpmul(mpw* result, size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata)
* \brief This function computes a full multi-precision product.
*/
BEECRYPTAPI
void mpmul(mpw* result, size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata);
/*!\fn void mpsqr(mpw* result, size_t size, const mpw* data)
* \brief This function computes a full multi-precision square.
*/
BEECRYPTAPI
void mpsqr(mpw* result, size_t size, const mpw* data);
BEECRYPTAPI
void mpgcd_w(size_t size, const mpw* xdata, const mpw* ydata, mpw* result, mpw* wksp);
BEECRYPTAPI
int mpextgcd_w(size_t size, const mpw* xdata, const mpw* ydata, mpw* result, mpw* wksp);
BEECRYPTAPI
mpw mppndiv(mpw xhi, mpw xlo, mpw y);
BEECRYPTAPI
void mpmod (mpw* result, size_t xsize, const mpw* xdata, size_t ysize, const mpw*ydata, mpw* wksp);
BEECRYPTAPI
void mpndivmod(mpw* result, size_t xsize, const mpw* xdata, size_t ysize, const mpw* ydata, mpw* wksp);
/*
* Output Routines
*/
BEECRYPTAPI
void mpprint(size_t size, const mpw* data);
BEECRYPTAPI
void mpprintln(size_t size, const mpw* data);
BEECRYPTAPI
void mpfprint(FILE* f, size_t size, const mpw* data);
BEECRYPTAPI
void mpfprintln(FILE* f, size_t size, const mpw* data);
/*
* Conversion Routines
*/
BEECRYPTAPI
int os2ip(mpw* idata, size_t isize, const byte* osdata, size_t ossize);
BEECRYPTAPI
int i2osp(byte* osdata, size_t ossize, const mpw* idata, size_t isize);
BEECRYPTAPI
int hs2ip(mpw* idata, size_t isize, const char* hsdata, size_t hssize);
#ifdef __cplusplus
}
#endif
#endif