h1-mod/deps/libtommath/doc/tommath.3

.TH LIBTOMMATH 3 "2003-28-02"
.SH NAME
libtommath - a big integer library
.SH SYNOPSIS
.sp
.ft B
.nf
#include <tommath\&.h>
.fi
.ft
.SH DESCRIPTION
.I LibTomMath
provides a series of efficient and carefully written functions
for manipulating large integer numbers.
.br
Functions are in alphabetical order.
.LP
\fBNOTE:\fP The errors listed are not the only ones possible, just the interesting ones.

.LP
.BI "mp_err mp_2expt(mp_int * " a ", int " b ")"
.in 1i
Computes 2^b with b >= 0 and puts the result in \fIa\fP. This functions uses a faster method than \fBmp_mul_2\fP.
.br
Returns \fBMP_OVF\fP if the result would be larger than a \fBmp_int\fP can hold. The macro \fBMP_MAX_DIGIT_COUNT\fP holds
the maximal number of limbs: ((INT_MAX - 2) / MP_DIGIT_BIT).
.br
Returns \fBMP_VAL\fP if b < 0
.in -1i


.LP
.BI "mp_err mp_abs (const mp_int *" a ", mp_int *" b ")"
.in 1i
Computes the absolute value of \fBa\fP and puts the result in \fBb\fP.
.in -1i

.LP
.BI "mp_err mp_add (const mp_int *" a ", const mp_int *" b ", mp_int *" c ")"
.in 1i
Computes \fBa + b = c\fP.
.in -1i

.LP
.BI "mp_err mp_add_d(const mp_int *" a ", mp_digit " b ", mp_int *" c ")"
.in 1i
Computes \fBa + b = c\fP where \fBb\fP is of type \fBmp_digit\fP.
.in -1i

.LP
.BI "mp_err mp_addmod(const mp_int *" a ", const mp_int *" b ", const mp_int *" c ", mp_int *" d ")"
.in 1i
Computes \fB(a + b) % c = d\fP. No optimizations.
.in -1i

.LP
.BI "mp_err mp_and (const mp_int *" a ", mp_int *" b ", mp_int *" c ")"
.in 1i
Computes bitwise or \fBa & b = c\fP. Negative numbers
are treated as if they are in two-complement representation.
.in -1i

.LP
.BI "void mp_clamp(mp_int *" a ");
.in 1i
This is used to ensure that leading zero digits are trimmed and the leading \fBused\fP digit will be
non-zero. It also fixes the sign if there are no more leading digits.
.in -1i

.LP
.BI "void mp_clear (mp_int *" a ")"
.in 1i
Frees the heap memory of \fBa\fP.
.in -1i

.LP
.BI "int mp_cnt_lsb(const mp_int *" a ")"
.in 1i
Returns the position of the lowest bit set.
.in -1i

.LP
.BI "mp_ord mp_cmp(const mp_int *" a ", const mp_int *" b ")"
.in 1i
Compare \fBa\fP to \fBb\fP.
.in -1i

.LP
.BI "mp_ord mp_cmp_d(const mp_int *" a ", mp_digit " b ")"
.in 1i
Compare \fBa\fP to a single digit \fBb\fP.
.in -1i

.LP
.BI "mp_ord mp_cmp_mag(const mp_int *" a ", const mp_int *" b ")"
.in 1i
Compares the absolute values of \fBa\fP and \fBb\fP.
.in -1i

.LP
.BI "mp_err mp_complement(const mp_int *" a ", mp_int *" b ")"
.in 1i
Computes the 2-complement \fIb = ~a\fP.
.in -1i

.LP
.BI "mp_err mp_copy (const mp_int *" a ", mp_int *" b ")"
.in 1i
Makes a deep copy of \fBa\fP into \fBb\fP.
.in -1i

.LP
.BI "int mp_count_bits(const mp_int *" a ")"
.in 1i
Returns the position of the highest bit set.
.in -1i

.LP
.BI "mp_err mp_decr(mp_int *" a ")"
.in 1i
Computes \fBa--\fP.
.in -1i

.LP
.BI "mp_err mp_div_2(const mp_int *" a ", mp_int *" b ");
.in 1i
Computes \fBa >> 1 = b\fP.
.in -1i

.LP
.BI "mp_err mp_div_2d (const mp_int *" a ", int " b ", mp_int *" c ", mp_int *" d ");
.in 1i
Computes \fBa >> b = c + d\fP with \fBc\fP the optional quotient and \fBd\fP the optional remainder.
.br
Set the argument for quotient and/or remainder to \fBNULL\fP to ignore the respective output.
.br
Returns \fBMP_VAL\fP if \fBb < 0\fP
.in -1i

.LP
.BI "mp_err mp_div (const mp_int *" a ", const mp_int *" b ", mp_int *" c ", mp_int *" d ")"
.in 1i
Computes  \fBa / b = c + d\fP with \fBc\fP the optional quotient and \fBd\fP the optional remainder.
.br
Set the argument for quotient and/or remainder to \fBNULL\fP to ignore the respective output.
.br
This function calls one of either: \fBs_mp_div_recursive\fP, \fBs_mp_div_school\fP, or
\fBs_mp_div_small\fP.
.br
Returns \fBMP_VAL\fP if \fBb = 0\fP
.in -1i

.LP
.BI "mp_err mp_div_d(const mp_int *" a ", mp_digit " b ", mp_int *" c ", mp_digit *" d ")"
.in 1i
Computes  \fBa / b = c + d\fP with \fBc\fP the optional quotient and \fBd\fP the optional remainder.
.br
Set the argument for quotient and/or remainder to \fBNULL\fP to ignore the respective output.
Returns \fBMP_VAL\fP if \fBb = 0\fP.
.in -1i

.LP
.BI "bool mp_dr_is_modulus(const mp_int *" a ")"
.in 1i
Returns \fBtrue\fP if the modulus \fBa\fP is of the form below that allows for a diminished radix reduction, \fBfalse\fP otherwise.
.br
\fB\[*b]^k - p\fP for some \fBk >= 0\fP and \fB0 < p < \[*b]\fP where \fB\[*b]\fP is the radix.
.in -1i

.LP
.BI "mp_err mp_dr_reduce(mp_int *" a ", const mp_int *" b ", mp_digit " mp ")"
.in 1i
This reduces \fBa\fP in place modulo \fBb\fP with the pre-computed value \fBmp\fP. \fBb\fP must be of a restricted
diminished radix form and \fBa\fP must be in the range \fB0 <= a < b^2\fP
.in -1i

.LP
.BI "void mp_dr_setup(const mp_int *" a ", mp_digit *" d ")"
.in 1i
This computes the value required for the modulus \fBa\fP and stores it in \fBd\fP.
.in -1i

.LP
.BI "void mp_exch (mp_int *" a ", mp_int *" b ")"
.in 1i
Swaps, but just the pointers.
.in -1i

.LP
.BI "const char *mp_error_to_string(mp_err " code ")"
.in 1i
Returns a short ASCII message describing the error code given in \fBcode\fP.
.in -1i

.LP
.BI "mp_err mp_exptmod (const mp_int *" G ", const mp_int *" X ", const mp_int *" P ", mp_int *" Y ")"
.in 1i
This computes \fBY \[==] G^X (mod P)\fP using a variable width sliding window
algorithm.  This function will automatically detect the fastest modular reduction technique to use
during the operation. For negative values of \fBX\fP the operation is performed as \fBY \[==] (G^-1 mod P)^(|X| (mod P))\fP
provided that \fBgcd(G, P) = 1\fP.
.br
This function is actually a shell around the two internal exponentiation functions.  This routine
will automatically detect when Barrett, Montgomery, Restricted and Unrestricted Diminished Radix
based exponentiation can be used.  Generally moduli of the a "restricted diminished radix" form
lead to the fastest modular exponentiations. Followed by Montgomery and the other two algorithms.
.br
Returns \fBMP_VAL\fP if \fBP < 0\fP.
.br
Returns \fBMP_VAL\fP if none of the underlying internal functions have been compiled in.
.in -1i

.LP
.BI "mp_err mp_expt_n(const mp_int *" a ", int " b ", int *" c ")"
.in 1i
Computes \fBa^b = c\fP. Simple binary exponentiation, no further optimizations.
.br
\fBb\fP must be positive.
.in -1i

.LP
.BI "mp_err mp_exteuclid(const mp_int *" a ", const mp_int *" b ", mp_int *" U1 ", mp_int *" U2 ", mp_int *" U3 ")"
.in 1i
Computes the extended Euclidean algorithm: \fBa * U1 + b * U2 = U3\fP
.br
Set the argument for \fBU1, U2, U3\fP to \fBNULL\fP to ignore the respective output.
.in -1i

.LP
.BI "mp_err mp_fread(mp_int *" a ", int " radix ", FILE *" stream ")"
.in 1i
Reads a number in radix \fBradix\fP from file \fBstream\fP and converts it into the big integer \fBa\fP.
.br
Returns \fBMP_VAL\fP if radix is not in the range \fB2 <= radix <= 64\fP.
.br
Returns \fBMP_ERR\fP if no digits were found in \fBstream\fP.
.in -1i

.LP
.BI "mp_err mp_from_sbin(mp_int *" a ", const uint8_t *" b ", size_t " size ")"
.in 1i
This will read in an big-endian array of octets from \fBb\fP of length
\fBsize\fP into \fBa\fP.
.br
If the first octet of the data is zero, the sign of the big-integer will be \fBMP_NEG\fP and \fBMP_ZPOS\fP otherwise.
.in -1i

.LP
.BI "mp_err mp_from_ubin(mp_int *" a ", uint8_t *" b ", size_t " size ")"
.in 1i
This will read in an big-endian array of octets from \fBb\fP of length
\fBsize\fP into \fBa\fP.  The resulting big-integer \fBa\fP will always be positive.
.in -1i

.LP
.BI "mp_err mp_fwrite(const mp_int *" a ", int " radix ", FILE *" stream ")"
.in 1i
Writes \fBa\fP as a string representing the big integer in radix \fBradix\fP to file \fBstream\fP.
.br
Returns \fBMP_MEM\fP if the functions fails to allocate memory for the buffer.
.br
Returns \fBMP_ERR\fP if there was a problem writing to the file.
.br
Returns \fBMP_VAL\fP if radix is not in the range \fB2 <= radix <= 64\fP.
.in -1i

.LP
.BI "mp_err mp_gcd (const mp_int *" a ", const mp_int *" b ", mp_int *" c ")"
.in 1i
Compute the greatest common divisor of \fBa\fP and \fBb\fP and store it in \fBc\fP.
.in -1i

.LP
.BI "double mp_get_double(const mp_int *" a ")"
.in 1i
Returns a float of type \fBdouble\fP (binary64).
.br
Will overflow if the big integer is too big.
.br
\fBNOTE:\fP rounding mode is not set just taken. Use e.g.: \fBfesetround(3)\fP in \fBfenv.h\fP to
change the rounding mode.
.in -1i

.LP
.BI "int32_t mp_get_i32 (const mp_int *" a ")"
.in 1i
Returns a signed 32-bit integer from big-integer \fBa\fP.
.br
\fBNOTE:\fP This group of functions is truncating. Example:
.br
.TS
tab(;), allbox;
l r.
Input;123456789101112131415161718192021222324252627282930
mp_get_i32;-1073632270
mp_get_i64;5543444065158278130
mp_get_mag_u32;3221335026
mp_get_mag_u64;5543444065158278130
mp_get_l;5543444065158278130
mp_get_ul;5543444065158278130
mp_get_mag_ul;5543444065158278130
.TE
.in -1i

.LP
.BI "int64_t mp_get_i64 (const mp_int *" a ")"
.in 1i
Returns a signed 64-bit integer from big-integer \fBa\fP.
.br
\fBNOTE:\fP This function is truncating. See \fBmp_get_i32\fP for details.
.in -1i

.LP
.BI "long mp_get_l (const mp_int *" a ")"
.in 1i
Returns a signed \fBlong\fP from big-integer \fBa\fP.
.br
\fBNOTE:\fP This function is truncating. See \fBmp_get_i32\fP for details.
.in -1i

.LP
.BI "uint32_t mp_get_mag_u32 (const mp_int *" a ")"
.in 1i
Returns an unsigned 32 bit integer from big-integer \fBa\fP.
.br
\fBNOTE:\fP This function is truncating. See \fBmp_get_i32\fP for details.
.in -1i

.LP
.BI "uint64_t mp_get_mag_u64 (const mp_int *" a ")"
.in 1i
Returns an unsigned 64 bit integer from big-integer \fBa\fP.
.br
\fBNOTE:\fP This function is truncating. See \fBmp_get_i32\fP for details.
.in -1i

.LP
.BI "unsigned long mp_get_mag_ul (const mp_int *" a ")"
.in 1i
Returns an unsigned long from big-integer \fBa\fP.
.br
\fBNOTE:\fP This function is truncating. See \fBmp_get_i32\fP for details.
.in -1i

.LP
.BI "uint32_t mp_get_u32 (const mp_int *" a ")"
.in 1i
Convenience macro for \fBmp_get_mag_u32()\fP.
.in -1i

.LP
.BI "uint64_t mp_get_u64 (const mp_int *" a ")"
.in 1i
Convenience macro for \fBmp_get_mag_u64()\fP.
.in -1i

.LP
.BI "unsigned long mp_get_ul (const mp_int *" a ")"
.in 1i
Convenience macro for \fBmp_get_mag_ul()\fP.
.in -1i

.LP
.BI "mp_err mp_grow (mp_int *" a ", int " size ")"
.in 1i
This will grow the array of digits of \fBa\fP to \fBsize\fP.
.br
Returns \fBMP_MEM\fP if the functions fails to allocate enough memory.
.in -1i

.LP
.BI "mp_err mp_hash (const mp_int *" a ", mp_hval *" hash ")"
.in 1i
This will create the hash of \fBa\fP following the \fIFNV-1a\fP algorithm as described on
\fIhttp://www.isthe.com/chongo/tech/comp/fnv/index.html#FNV-1a\fP. With the
help of this function one can use an \fBmp_int\fP as a key in a hash table.
.br
\fBNOTE:\fP The hashing is not stable over different widths of a \fBmp_digit\fP.
.in -1i

.LP
.BI "mp_err mp_incr(mp_int *" a ")"
.in 1i
Computes \fBa++\fP.
.br
Returns \fBMP_MEM\fP if the reallocation of memory in \fBmp_grow\fP failed.
.in -1i

.LP
.BI "mp_err mp_init_copy (mp_int *" a ", mp_int *" b ")"
.in 1i
Initializes \fBa\fP and copies \fBb\fP into \fBa\fP.
.in -1i

.LP
.BI "mp_err mp_init_i32 (mp_int *" a ", int32_t " b ");
.in 1i
Initializes \fBa\fP and copies the signed 32 bit integer \fBb\fP into \fBa\fP.
.in -1i

.LP
.BI "mp_err mp_init_i64 (mp_int *" a ", int64_t " b ")"
.in 1i
Initializes \fBa\fP and copies the signed 64 bit integer \fBb\fP into \fBa\fP.
.in -1i

.LP
.BI "mp_err mp_init_l (mp_int *" a ", long " b ")"
.in 1i
Initializes \fBa\fP and copies the signed integer \fBb\fP of type \fBlong\fP into \fBa\fP.
.in -1i

.LP
.BI "mp_err mp_init (mp_int *" a ")"
.in 1i
Initializes \fBa\fP.
.br
It allocates a certain amount of memory such that the \fBmp_set_u64\fP setter can
store an \fBuint64_t\fP in the \fBmp_int\fP. To be able to take advantage of the algorithm used for
\fBmp_school_div\fP the \fBmp_int\fP must have at least three limbs.
.br
It actively sets all limbs to zero, overwriting what was there before, sets the \fBa.sign\fP to \fBMP_ZPOS\fP,
and \fBa.used\fP to zero.
.in -1i

.LP
.BI "mp_err mp_init_multi(mp_int *" mp ", " ... ")"
.in 1i
Initialize a \fBNULL\fP terminated series of \fBmp_int\fPs.
.in -1i

.LP
.BI "mp_err mp_init_set (mp_int *" a ", mp_digit " b ")"
.in 1i
Initializes \fBa\fP and sets it to the \fBmp_digit\fP \fBb\fP
.in -1i

.LP
.BI "mp_err mp_init_size (mp_int *" a ", int " size ")"
.in 1i
Initializes \fBa\fP with pre-grown to a \fBsize\fP number of limbs.
.br
If \fBsize\fP is smaller than the minimum (see \fBmp_init\fP for details)  \fBsize\fP will be increased to that minimum.
.br
Returns \fBMP_OVF\fP if \fBsize\fP is larger than an \fBmp_int\fP is able to hold.
.in -1i

.LP
.BI "mp_err mp_init_u32 (mp_int *" a ", uint32_t " b ")"
.in 1i
Initializes \fBa\fP and copies the unsigned 32 bit integer \fBb\fP into \fBa\fP.
.in -1i

.LP
.BI "mp_err mp_init_u64 (mp_int *" a ", uint64_t " b ")"
.in 1i
Initializes \fBa\fP and copies the unsigned 64 bit integer \fBb\fP into \fBa\fP.
.in -1i

.LP
.BI "mp_err mp_init_ul (mp_int *" a ", unsigned long " b ")"
.in 1i
Initializes \fBa\fP and copies the unsigned integer \fBb\fP of type \fBunsigned long\fP into \fBa\fP.
.in -1i

.LP
.BI "mp_err mp_invmod (const mp_int *" a ", const mp_int *" b ", mp_int *" c ")"
.in 1i
Computes the multiplicative inverse of \fBa\fP modulo \fBb\fP and stores the result in \fBv\fP such that
\fBac \[==] 1 (mod b)\fP.
.br
Does use a faster algorithm if the modulus \fBb\fP is odd.
.br
Returns \fBMP_VAL\fP if \fBb <= 1\fP
.in -1i

.LP
.BI "bool mp_iseven(const mp_int *" a ")"
.in 1i
Returns \fBtrue\fP if \fBa\fP is even, \fBfalse\fP otherwise.
.in -1i

.LP
.BI "bool mp_isneg(mp_int *" a ")"
.in 1i
Returns \fBtrue\fP if \fBa < 0\fP, \fBfalse\fP otherwise.
.in -1i

.LP
.BI "bool mp_isodd(const mp_int *" a ")"
.in 1i
Returns \fBtrue\fP if \fBa\fP is odd, \fBfalse\fP otherwise.
.in -1i

.LP
.BI "mp_err mp_is_square(const mp_int *" arg ", bool *" ret ")"
.in 1i
Sets \fBret\fP to \fBtrue\fP if \fBarg\fP is a square, \fBfalse\fP otherwise.
.in -1i

.LP
.BI "bool mp_iszero(mp_int *" a ")"
.in 1i
Returns \fBtrue\fP if \fBa = 0\fP, \fBfalse\fP otherwise.
.in -1i

.LP
.BI "bool mp_isone(mp_int *" a ")"
.in 1i
Returns \fBtrue\fP if \fBa = 1\fP, \fBfalse\fP otherwise.
.in -1i

.LP
.BI "mp_err mp_kronecker (const mp_int *" a ", const mp_int *" p ", int *" c ")"
.in 1i
Computes the Kronecker symbol (an extension of the Jacobi symbol) for \fBa\fP with respect to
\fBp\fP with \fB(a, p) in Z\fP.  If \fBp\fP is prime this essentially computes the
Legendre symbol. The result is stored in \fBc\fP and can take on one of three values \fB{-1, 0, 1}\fP.
.TS
tab(;);
r l.
-1 ;if \fBa\fP is not a quadratic residue modulo \fBp\fP and \fBp\fP is prime.
0 ;if \fBa\fP divides \fBp\fP
1 ;if \fBa\fP is a quadratic residue modulo \fBp\fP.
.TE
.in -1i

.LP
.BI "mp_err mp_lcm (const mp_int *" a ", const mp_int *" b ", mp_int *" c ")"
.in 1i
Computes the least common multiple as \fB|a * b|/gcd(a, b)\fP.
.in -1i

.LP
.BI "mp_err mp_log(const mp_int *" a ", const mp_int *" base ", int *" c ")"
.in 1i
Computes \fBlog_b(a)\fP such that \fB(log_b a)^b <= a\fP.
.br
Returns \fBMP_VAL\fP if \fBa <= 0\fP or \fBb < 2\fP.
.in -1i

.LP
.BI "mp_err mp_log_n(const mp_int *" a ", int " base ", int *" c ")"
.in 1i
Computes \fBlog_b(a)\fP such that \fB(log_b a)^b <= a\fP.
.br
Convenience function that is a wrapper for \fBmp_log\fP doing the conversion of \fBb\fP
to a big integer.
.in -1i

.LP
.BI "mp_err mp_lshd (mp_int *" a ", int " b ")"
.in 1i
Shift \fBa\fP left by \fBb\fP limbs: \fBa * 2^(b*MP_DIGIT_BIT)\fP.
.in -1i

.LP
.BI "mp_err mp_mod_2d(const mp_int *" a ", int " b ", mp_int *" c ")"
.in 1i
Compute \fBa % 2^b\fP.
.br

.in -1i

.LP
.BI "mp_err mp_mod(const mp_int *" a ",const mp_int *" b ", mp_int *" c ")"
.in 1i
Compute \fBa \[==] c mod b\fP.
.br
\fB0 <= c < b\fP if \fBb > 0\fP and \fBb < c <= 0\fP if \fBb < 0\fP.
.in -1i

.LP
.BI "mp_err mp_mod_d(const mp_int *" a ", mp_digit " b ", mp_digit *" c ")"
.in 1i
Computes the remainder of \fBa / b\fP.
.in -1i

.LP
.BI "mp_err mp_montgomery_calc_normalization(mp_int *" a ", mp_int *" b ")"
.in 1i
Computes \fBR = r^n\fP for Montgomery reduction where \fBn\fP is size of \fBa\fP in bits and
\fBr\fP is the radix (\fBMP_DIGIT_MAX + 1\fP).
.br
See \fBmp_montgomery_reduce\fP for some example code.
.in -1i

.LP
.BI "mp_err mp_montgomery_reduce(mp_int *" a ", mp_int *" m ", mp_digit " mp ")"
.in 1i
Reduces \fBa\fP in place modulo \fBm\fP with the pre-computed value \fBmp\fP (\fBa*mp^(-1) \[==] x (mod m) \fP).
Pre-computation of \fBmp\fP can be done with \fBmp_montgomery_setup\fP. Example:

.nf
int main(void)
{
   mp_int   a, b, c, R;
   mp_digit mp;
   mp_err      result;

   /* initialize a,b to desired values,
    * mp_init R, c and set c to 1....
    */

   /* get normalization */
   if ((result = mp_montgomery_calc_normalization(&R, b)) != MP_OKAY) {
      printf("Error getting norm.  %s",
             mp_error_to_string(result));
      return EXIT_FAILURE;
   }
   /* get mp value */
   if ((result = mp_montgomery_setup(&c, &mp)) != MP_OKAY) {
      printf("Error setting up montgomery.  %s",
             mp_error_to_string(result));
      return EXIT_FAILURE;
   }

   /* normalize `a' so now a is equal to aR */
   if ((result = mp_mulmod(&a, &R, &b, &a)) != MP_OKAY) {
      printf("Error computing aR.  %s",
             mp_error_to_string(result));
      return EXIT_FAILURE;
   }

   /* square a to get c = a^2R^2 */
   if ((result = mp_sqr(&a, &c)) != MP_OKAY) {
      printf("Error squaring.  %s",
             mp_error_to_string(result));
      return EXIT_FAILURE;
   }
   /* now reduce `c' back down to c = a^2R^2 * R^-1 == a^2R */
   if ((result = mp_montgomery_reduce(&c, &b, mp)) != MP_OKAY) {
      printf("Error reducing.  %s",
             mp_error_to_string(result));
      return EXIT_FAILURE;
   }

   /* multiply a to get c = a^3R^2 */
   if ((result = mp_mul(&a, &c, &c)) != MP_OKAY) {
      printf("Error reducing.  %s",
             mp_error_to_string(result));
      return EXIT_FAILURE;
   }

   /* now reduce `c' back down to c = a^3R^2 * R^-1 == a^3R */
   if ((result = mp_montgomery_reduce(&c, &b, mp)) != MP_OKAY) {
      printf("Error reducing.  %s",
             mp_error_to_string(result));
      return EXIT_FAILURE;
   }

   /* now reduce (again) `c' back down to c = a^3R * R^-1 == a^3 */
   if ((result = mp_montgomery_reduce(&c, &b, mp)) != MP_OKAY) {
      printf("Error reducing.  %s",
             mp_error_to_string(result));
      return EXIT_FAILURE;
   }

   /* c now equals a^3 mod b */

   return EXIT_SUCCESS;
\}
.in -1i

.LP
.BI "mp_err mp_montgomery_setup(const mp_int *" a ", mp_digit *" mp ")"
.in 1i
For the given odd modulus \fBa\fP the pre-computation value is placed in \fBmp\fP.
.br
\fBmp = 1/a mod 2^k\fP with \fBk\fP the number of bits in the underlying native type used for \fBmp_digit\fP.
.br
See \fBmp_montgomery_reduce\fP for some example code.
.in -1i

.LP
.BI "mp_err mp_mul_2(const mp_int *" a ", mp_int *" b ")"
.in 1i
Computes \fBb = 2 * a\fP.
.in -1i

.LP
.BI "mp_err mp_mul_2d(const mp_int *" a ", int " b ", mp_int *" c ")"
.in 1i
Shifts \fBa\fP left by \fBb\fP bits \fBb = a * 2^b\fP.
.br
Condition: \fBb >= 0\fP
.in -1i

.LP
.BI "mp_err mp_mul (const mp_int *" a ", const mp_int *" b ", mp_int *" c ")"
.in 1i
Computes \fBc = a * b\fP.
.in -1i

.LP
.BI "mp_err mp_mul_d(const mp_int *" a ", mp_digit " b ", mp_int *" c ")"
.in 1i
Computes \fBc = a * b\fP with \fBb\fP and \fBmp_digit\fP.
.in -1i

.LP
.BI "mp_err mp_mulmod(const mp_int *" a ", const mp_int *" b ", const mp_int *" c ", mp_int *" d ")"
.in 1i
Computes \fBd = a * b (mod c)\fP. No optimizations.
.br
For special forms of the input libtommath offers other, optimized algorithms. See for example
\fBmp_montgomery_reduce\fP for some example code with Montgomery reduction. There is also Barret
reduction, which is more generic (\fBmp_reduce\fP), diminished radix reduction (\fBmp_dr_reduce\fP) and
unrestricted diminished radix reduction (\fBmp_reduce_2k\fP)
.br
.in -1i

.LP
.BI "mp_err mp_neg (const mp_int *" a ", mp_int *" b ")"
.in 1i
Computes \fBb = -a\fP.
.in -1i

.LP
.BI "mp_err mp_or  (const mp_int *" a ", mp_int *" b ", mp_int *" c ")"
.in 1i
Computes bit-wise or \fBa | b = c\fP. Negative numbers
are treated as if they are in two-complement representation.
.in -1i

.LP
.BI "size_t mp_pack_count(const mp_int *" a ", size_t " nails ", size_t " size ")"
.in 1i
Returns the size in bytes necessary to be put in \fBmp_pack\fP's \fBmaxsize\fP. See \fBmp_pack\fP for details.
.in -1i

.LP
.BI "mp_err mp_pack(void *" rop ", size_t *" countp ", mp_order " order ", size_t " size ", mp_endian " endian ", size_t " nails ", const mp_int *" op ")"
.in 1i
Export binary data.
.br
Implements the similarly working GMP functions as described at \fIhttp://gmplib.org/manual/Integer-Import-and-Export.html\fP with
the exception that \fBmp_pack\fP will not allocate memory if \fBrop\fP is \fBNULL\fP.
.br
To make things a bit more comfortable libtommath offers two \fBenum\fPs:

.br
.in 1.5i
typedef enum {
   MP_LSB_FIRST = -1,
   MP_MSB_FIRST =  1
} mp_order;

.br

typedef enum {
   MP_LITTLE_ENDIAN  = -1,
   MP_NATIVE_ENDIAN  =  0,
   MP_BIG_ENDIAN     =  1
} mp_ndian;
.in -1.5i
.in -1i

.LP
.BI "mp_err mp_prime_fermat(const mp_int *" a ", const mp_int *" b ", bool *" result ")"
.in 1i
Performs one Fermat test of \fBa\fP using base \fBb\fP.
.br
Sets \fBresult\fP to \fBtrue\fP if \fBa\fP is probably prime, \fBfalse\fP if composite.
.in -1i

.LP
.BI "mp_err mp_prime_strong_lucas_selfridge(const mp_int *" a ", bool *" result ")"
.in 1i
Sets \fBresult\fP to \fBtrue\fP if \fBa\fP is a strong Lucas-Selfridge pseudoprime, \fBfalse\fP otherwise.
.br
It has been verified that this function together with one round of Miller-Rabin to the base 2 (two) is
deterministic up to 2^64.
.in -1i

.LP
.BI "mp_err mp_prime_frobenius_underwood(const mp_int *" N ", bool *" result ")"
.in 1i
Sets \fBresult\fP to \fBtrue\fP if \fBa\fP is a Frobenius-Underwood pseudoprime, \fBfalse\fP otherwise.
.br
It has been verified that this function (as a stand-alone) is deterministic up to at least 2^50.
.in -1i

.LP
.BI "mp_err mp_prime_is_prime(const mp_int *" a ", int " t ", bool *" result ")"
.in 1i
Sets \fBresult\fP to \fBtrue\fP if \fBa\fP is a pseudoprime, \fBfalse\fP otherwise.
.br
The argument \fBt\fP holds the number of random Miller-Rabin tests to be executed.
.br
.nr step 1 1
Uses a stack of different tests to detect composites:
.RS
.IP \n[step]  0.3i
Direct test: checks if input is one of the values 0, 1, 2.
.IP \n+[step]
The only even prime has been handled, reject even input from now on
.IP \n+[step]
Check if input is a square. (some of the algorithms used later do not like square input)
.IP \n+[step]
Check if input is equal to one of the primes in the table \fBs_mp_prime_tab\fP
.IP \n+[step]
Check if input is divisible by one of the primes in the table \fBs_mp_prime_tab\fP
.IP \n+[step]
Run a Miller-Rabin test with base 2
.IP \n+[step]
Run a Miller-Rabin test with base 3
.IP \n+[step]
If \fBt <= 0\fP and the macro \fBLTM_USE_ONLY_MR\fP is not defined we run either a Frobenius-Underwood
test if the macro \fBLTM_USE_FROBENIUS_TEST\fP is defined or a Lucas-Selfridge test.
.br
The Lucas-Selfridge test together with the two Miller-Rabin tests earlier is deterministic up to at least 2^64.
.br
The Frobenius-Underwood test as a stand-alone is deterministic up to at least 2^50. But it is different from
the Lucas-Selfridge test so the additional cost (about two times the time a Lucas-Selfridge test would need) might be worthwhile.
.IP \n+[step]
Even if \fBt = 0\fP we run at least one Miler-Rabin test with a random base.
.IP \n+[step]
If \fBt < 0\fP and input smaller than 3,317,044,064,679,887,385,961,981 (< 82 bits) several Miller-Rabin tests
are run with bases according to Sorenson, Jonathan; Webster, Jonathan (2015) "Strong Pseudoprimes to Twelve Prime Bases".
.br
Here ends the deterministic part of this function.
.IP \n+[step]
If \fBt > 0\fP: run the given number of Miller-Rabin tests with random bases.
.RE
.in -1i

.LP
.BI "mp_err mp_prime_miller_rabin (const mp_int *" a ", const mp_int *" b ", int *" result ")"
.in 1i
Run the Miller-Rabin pseudoprime test of \fBa\fP with base \fBb\fP and set \fBresult\fP to \fBtrue\fP
if \fBa\fP is a strong pseudprime for base \fBb\fP and \fBfalse\fP otherwise.
.in -1i

.LP
.BI "mp_err mp_prime_next_prime(mp_int *" a ", int " t ", bool " bbs_style ")"
.in 1i
Sets \fBa\fP to the next prime, even if \fBa\fP is prime itself.
.br
Argument \fBt\fP holds the number of Miller-Rabin tests to random bases and can also be used to steer
\fBmp_prime_is_prime\fP as described there.
.br
Argument \fBbbs_style\fP returns only primes \fBa \[==] 3 (mod 4)\fP if set to \fBtrue\fP.
.in -1i

.LP
.BI "mp_err mp_prime_rabin_miller_trials(int " size ")"
.in 1i
Returns the number of Miller-Rabin tests to random bases necessary for RSA according to
FIPS 186-4.
.br
\fBsize\fP is the size of the prime in bits.
.br
The entries are pre-computed:
.br
.TS
tab(;), allbox;
c c c
r r l.
\fB<=size\fP;\fB#tests\fP ;\fBError\fP
.SP
 80;-1;Use deterministic algorithm for size <= 80 bits
 81;37;2^(-96)
 96;32;2^(-96)
 128;40;2^(-112)
 160;35;2^(-112)
 256;27;2^(-128)
 384;16;2^(-128)
 512;18;2^(-160)
 768;11;2^(-160)
 896;10;2^(-160)
 1024;12;2^(-192)
 1536;8;2^(-192)
 2048;6;2^(-192)
 3072;4;2^(-192)
 4096;5;2^(-256)
 5120;4;2^(-256)
 6144;4;2^(-256)
 8192;3;2^(-256)
 9216;3;2^(-256)
 10240;2;For bigger keysizes use always at least 2 Rounds
.TE
.in -1i

.LP
.BI "mp_err mp_prime_rand(mp_int *" a ", int " t ", int " size ", int " flags ")"
.in 1i
Generates a random big prime \fBa\fP with \fBsize\fP bits.
.br
Argument \fBt\fP holds the number of Miller-Rabin tests to random bases and can also be used to steer
\fBmp_prime_is_prime\fP as described there.
.br
Argument \fBflags\fP determines the kind of prime:
.br
.TS
tab(;), allbox;
c c
l l.
\fBFlag\fP;\fBCondition\fP
MP_PRIME_BBS;make prime congruent to 3 mod 4
MP_PRIME_SAFE;make sure (p-1)/2 is prime as well (implies MP_PRIME_BBS)
MP_PRIME_2MSB_ON;make the 2nd highest bit one
.TE
.in -1i

.LP
.BI "mp_err mp_radix_size (const mp_int *" a ", int " radix ", int *" size ")"
.in 1i
Sets \fBsize\fP to the length of the ASCII string of the representation of big integer \fBa\fP in radix \fBradix\fP.
This includes the sign and the terminatong \fBNUL\fP.
.br
Returns \fBMP_VAL\fP if \fBradix\fP is not in the range \fB2 <= radix <= 64\fP
.in -1i

.LP
.BI "mp_err mp_radix_size_overestimate (const mp_int *" a ", int " radix ", int *" size ")"
.in 1i
Same as \fBmp_radix_size\fP but uses a rough, table based approximation instead of calling \fBmp_log_n\fP.
.br
Overestimates the result by some units: relative error is about \fB10^(-8)\fP. Experiments showed that
the absolute error should not go pass \fB5\fP.
.in -1i

.LP
.BI "mp_err mp_rand(mp_int *" a ", int " digits ")"
.in 1i
Generates a random big integer \fBa\fP with a \fBdigits\fP number of limbs.
.br
\fBNOTE:\fP This function uses the same (P)RNG as the prime generating/testing functions. If that entropy
is precious see \fBmp_rand_source\fP to set another (P)RNG.
.in -1i

.LP
.BI "void mp_rand_source(mp_err(*" source ")(void *" out ", size_t " size "));
.in 1i
Sets the (P)RNG used for \fIall\fP functions in Libtommath, that need some random bytes.
.br
Set \fBsource\fP to \fBNULL\fP to get the original (Libtommath) source back.
.br
Example:
.br
.in +.5i
.nf
uint32_t prng_state = 0xdeadbeef;
uint32_t bad_prng(void) {
   prng_state = (1103515245ul * prng_state  + 12345ul) % 2147483648ul;
   return prng_state;
}
mp_err myprng(void *p, size_t n)
{
   char *q = (char *)p;
   while (n > 0u) {
      int i;
      uint32_t x = bad_prng();
      for (i = 0; (i < 4) && (n > 0u); ++i, --n) {
         *q++ = (char)(x & 0xFFu);
         x >>= 8;
      }
   }
   return MP_OKAY;
}
mp_err some_monte_carlo_function(...)
{
   /* Use the the bad but fast P(R)NG */
   mp_rand_source(myprng);
   ...
   while(big_number--) {
      mp_rand(a_bigint, n_bits);
      ...
   }
   /* Reset and use (P)RNG provided by the platform */
   mp_rand_source(NULL);
   ...
}
.in -.5i
.in -1i

.LP
.BI "mp_err mp_read_radix (mp_int *" a ", const char *" str ", int " radix ")"
.in 1i
This will read a \fBNUL\fP terminated string in base \fBradix\fP from \fBstr\fP into \fBa\fP.
.br
Returns \fBMP_VAL\fP if \fBradix\fP is not in the range \fB2 <= radix <= 64\fP.
.in -1i

.LP
.BI "mp_err mp_reduce_2k_l(mp_int *" a ", const mp_int *" n ", const mp_int *" d ")"
.in 1i
Reduces \fBa\fP modulo \fBn\fP where \fBn\fP is of the form \fB2^p - d\fP where \fBd\fP
is a big integer.
.br
Belongs to the "unrestricted diminished radix reduction" method.
.in -1i

.LP
.BI "mp_err mp_reduce_2k(mp_int *" a ", const mp_int *" n ", mp_digit " d ")"
.in 1i
Reduces \fBa\fP modulo \fBn\fP where \fBn\fP is of the form \fB2^p - d\fP where \fBd\fP
is a \fBmp_digit\fP.
.br
Belongs to the "unrestricted diminished radix reduction" method.
.in -1i

.LP
.BI "mp_err mp_reduce_2k_setup(const mp_int *" a ", mp_digit *" d ")"
.in 1i
This will compute the required \fBd\fP value in \fB2^p - d\fP for the given moduli \fBa\fP where \fBd\fP
is a \fBmp_digit\fP.
.br
Belongs to the "unrestricted diminished radix reduction" method.
.in -1i

.LP
.BI "mp_err mp_reduce_2k_setup_l(const mp_int *" a ", mp_int *" d ")"
.in 1i
This will compute the required \fBd\fP value in \fB2^p - d\fP for the given moduli \fBa\fP where \fBd\fP
is a big integer.
.br
Belongs to the "unrestricted diminished radix reduction" method.
.in -1i

.LP
.BI "mp_err mp_reduce(const mp_int *" a ", const mp_int *" b ", mp_int *" c ")"
.in 1i
This will reduce \fBa\fP in place modulo \fBb\fP with the precomputed \fBmu\fP value in \fBc\fP.  \fBa\fPmust be in
the range \fB0 <= a < b^2\fP.
.br
Belongs to the "Barrett reduction" method.
.in -1i

.LP
.BI "bool mp_reduce_is_2k(const mp_int *" a ")"
.in 1i
Determines if \fBmp_reduce_2k\fP can be used with a \fBmp_digit\fP.
.br
Belongs to the "unrestricted diminished radix reduction" method.
.in -1i

.LP
.BI "bool mp_reduce_is_2k_l(const mp_int *" a ")"
.in 1i
Determines if \fBmp_reduce_2k\fP can be used with a big integer.
.br
Belongs to the "unrestricted diminished radix reduction" method.
.in -1i

.LP
.BI "mp_err mp_reduce_setup(const mp_int *" a ", mp_int *" b ")"
.in 1i
Compute \fBmu\fP for Barrett reduction.
.in -1i

.LP
.BI "mp_err mp_root_n(const mp_int *" a ", int " b ", mp_int *" c ")"
.in 1i
This computes \fBc = a^(1/b)\fP such that \fBc^b <= a\fP and \fB(c+1)^b > a\fP. Will return a positive root
only for even roots and return a root with the sign of the input for odd roots.  For example,
performing \fB4^(1/2)\fP will return \fB2\fP whereas \fB(-8)^(1/3)\fP will return \fB-2\fP.
.br
Returns \fBMP_VAL\fP if \fBb < 0\fP or \fBb > MP_DIGIT_MAX\fP or when \fBb\fP is even and \fBa < 0\fP.
.in -1i

.LP
.BI "void mp_rshd (mp_int *" a ", int " b ")"
.in 1i
Shift \fBa\fP right by \fBb\fP limbs: \fBa / 2^(b*MP_DIGIT_BIT)\fP.
.in -1i

.LP
.BI "size_t mp_sbin_size(const mp_int *" a ")"
.in 1i
Returns the number of bytes (octets) required to store the signed copy of the integer \fBa\fP.
.in -1i

.LP
.BI "mp_err mp_set_double(mp_int *" a ", double " b ")"
.in 1i
If the platform supports the floating point data type \fBdouble\fP (binary64) this function will
assign the integer part of \fBb\fP to the big integer \fBa\fP.
.br
Returns \fBMP_VAL\fP if \fBb\fP is \fB+/-infinity\fP or \fBNaN\fP
.in -1i

.LP
.BI "void mp_set_i32 (mp_int *" a ", int32_t " b ")"
.in 1i
Sets big integer \fBa\fP to the value of the signed 32 bit integer \fBb\fP.
.in -1i

.LP
.BI "void mp_set_i64 (mp_int *" a ", int64_t " b ")"
.in 1i
Sets big integer \fBa\fP to the value of the signed 64 bit integer \fBb\fP.
.in -1i

.LP
.BI "void mp_set_l (mp_int *" a ", long " b ")"
.in 1i
Sets big integer \fBa\fP to the value of the integer \fBb\fP of type \fIsigned long\fP.
.in -1i

.LP
.BI "void mp_set (mp_int *" a ", mp_digit " b ")"
.in 1i
Sets big integer \fBa\fP to the value of the integer \fBb\fP of type \fImp_digit\fP.
.in -1i

.LP
.BI "void mp_set_u32 (mp_int *" a ", uint32_t " b ")"
.in 1i
Sets big integer \fBa\fP to the value of the unsigned 32 bit integer \fBb\fP.
.in -1i

.LP
.BI "void mp_set_u64 (mp_int *" a ", uint64_t " b ")"
.in 1i
Sets big integer \fBa\fP to the value of the unsigned 64 bit integer \fBb\fP.
.in -1i

.LP
.BI "void mp_set_ul (mp_int *" a ", unsigned long " b ")"
.in 1i
Sets big integer \fBa\fP to the value of the integer \fBb\fP of type \fIunsigned long\fP.
.in -1i

.LP
.BI "mp_err mp_shrink (mp_int *" a ")"
.in 1i
This will remove excess digits of the \fImp_int\fP \fBa\fP.
.br
This method uses \fBrealloc\fP with all its problems, use with caution.
.in -1i

.LP
.BI "mp_err mp_signed_rsh(const mp_int *" a ", int " b ", mp_int *" c ")"
.in 1i
Shifts right \fBa\fP by \fBb\fP bits with sign extension.
.in -1i

.LP
.BI "mp_err mp_sqr (const mp_int *" a ", mp_int *" b ")"
.in 1i
Computes \fBa^2 = b\fP
.in -1i

.LP
.BI "mp_err mp_sqrmod(const mp_int *" a ", const mp_int *" b ", const mp_int *" c ")"
.in 1i
Computes \fBa^2 % b = c\fP. No optimization.
.in -1i

.LP
.BI "mp_err mp_sqrt(const mp_int *" arg ", mp_int *" ret ")"
.in 1i
Computes \fBa^(1/2) = b\fP such that \fBb^2 <= a\fP.
.in -1i

.LP
.BI "mp_err mp_sqrtmod_prime(const mp_int *" n ", const mp_int *" p ", mp_int *" r ")"
.in 1i
Computes\fBa^(1/2) % p = r\fP with \fBp\fP prime. Uses the Tonelli-Shanks algorithm.
.br
Does do some checks for \fBp\fP but no actual primality test.
.br
The prime \fBp\fP must be odd.
.in -1i

.LP
.BI "mp_err mp_sub (const mp_int *" a ", const mp_int *" b ", mp_int *" c ")"
.in 1i
Computes \fBa - b = c\fP with \fBb\fP a big integer.
.in -1i

.LP
.BI "mp_err mp_sub_d(const mp_int *" a ", mp_digit " b ", mp_int *" c ")"
.in 1i
Computes \fBa - b = c\fP with \fBb\fP of type \fImp_digit\fP.
.in -1i

.LP
.BI "mp_err mp_submod(const mp_int *" a ", const mp_int *" b ", const mp_int *" c ", mp_int *" d ")"
.in 1i
Computes \fB(a - b) % c = d\fP. No optimization.
.in -1i

.LP
.BI "mp_to_binary(" M ", " S ", " N ")"
.in 1i
Macro that calls \fBmp_to_radix\fP with fixed \fBradix = 2\fP.
.in -1i

.LP
.BI "mp_to_decimal(" M ", " S ", " N ")"
.in 1i
Macro that calls \fBmp_to_radix\fP with fixed \fBradix = 10\fP.
.in -1i

.LP
.BI "mp_to_hex(" M ", " S ", " N ")"
.in 1i
Macro that calls \fBmp_to_radix\fP with fixed \fBradix = 16\fP.
.in -1i

.LP
.BI "mp_to_octal(" M ", " S ", " N ")"
.in 1i
Macro that calls \fBmp_to_radix\fP with fixed \fBradix = 8\fP.
.in -1i

.LP
.BI "mp_err mp_to_radix (const mp_int *" a ", char *" str ", size_t " maxlen ", size_t *" written ", int " radix ")"
.in 1i
Stores upto \fBsize - 1\fP chars and always a \fBNULL\fP byte in \fBstr\fP and puts the number of characters written,
including \fBNUL\fP, in \fBwritten\fP.
.br
The caller is responsible to allocate a sufficient amount of memory for the buffer \fBstr\fP. Use \fBmp_radix_size\fP or
\fBmp_radix_size_overestimate\fP to have that number computed.
.br
Returns \fBMP_BUF\fP if \fBmaxlen < 2\fP
.br
Return \fBMP_VAL\fP if \fBradix\fP is not in the range \fB2 <= radix <= 64\fP.
.in -1i

.LP
.BI "mp_err mp_to_sbin(const mp_int *" a ", uint8_t *" b ", size_t " maxsize ", size_t *" len ")"
.in 1i
Store a \fBmaxlen\fP amount of the big integer in buffer \fBb\fP and puts the number of bytes written in \fBlen\fP.
.br
If \fBa\fP is negative \fBb0] = 1\fP and \fBb[0] = 0\fP otherwise.
.br
Returns \fBMP_BUF\fP if \fBmaxlen\fP is too small. Use \fBmp_sbin_size\fP to have that number computed.
.in -1i

.LP
.BI "mp_err mp_to_ubin(const mp_int *" a ", uint8_t *" buf ", size_t " maxlen ", size_t *" written ")"
.in 1i
Store a \fBmaxlen\fP amount of the big integer in buffer \fBb\fP and puts the number of bytes written in \fBlen\fP.
.br
Returns \fBMP_BUF\fP if \fBmaxlen\fP is too small. Use \fBmp_sbin_size\fP to have that number computed.
.in -1i

.LP
.BI "size_t mp_ubin_size(const mp_int *" a ")"
.in 1i
Returns the number of bytes (octets) required to store the unsigned copy of the big integer \fBa\fP.
.in -1i

.LP
.BI "mp_err mp_unpack(mp_int *" rop ", size_t " count ", mp_order " order ", size_t " size ", mp_endian " endian ", size_t " nails ", const void *" op ")"
.in 1i
Implements the similarly working GMP functions as described at \fIhttp://gmplib.org/manual/Integer-Import-and-Export.html\fP.
See also: \fBmp_pack\fP for more details.
.in -1i

.LP
.BI "mp_err mp_xor (const mp_int *" a ", mp_int *" b ", mp_int *" c ")"
.in 1i
Computes bit-wise xor \fBa ^ b = c\fP. Negative numbers are treated as if they are in two-complement representation.
.br
Returns \fBMP_MEM\fP if the reallocation of memory in \fBmp_grow\fP failed.
.in -1i

.LP
.BI "void mp_zero(mp_int *" a ")"
.in 1i
Sets \fBa\fP to zero.
.in -1i

.LP
.SS "TYPES"
.\"TODO: sort alphabetically
.BI mp_digit
.in 1i
An unsigned machine integer able to hold \fBMP_DIGIT_BIT + 1\fP bits.
.br
This type is defined in \fBtommath.h\fP
.in -1i
.LP
.BI mp_word
.in 1i
An unsigned integer able to hold a value of size \fB2 * MP_DIGIT_MAX + 1\fP.
This type is defined in \fBtommath_private.h\fP
.in -1i
.LP
.BI mp_int
.in 1i
The internal structure of Libtommath's big integer type \fBmp_int\fP.

.in 1.5i
.nf
typedef struct  {
   int used, alloc;
   mp_sign sign;
   mp_digit *dp;
} mp_int;
.in -1.5i
.in -1i
.LP
.BI mp_sign
.in 1i
The sign of Libtommath's big integer type \fBmp_int\fP

.in 1.5i
.nf
typedef enum {
   MP_ZPOS = 0,   /* positive */
   MP_NEG = 1     /* negative */
} mp_sign;
.in -1.5i
.in -1i

.LP
.BI mp_ord
.in 1i
Results of comparing with the functions \fBmp_cmp()\fP and  \fBmp_cmp_d()\fP.

.in 1.5i
.nf
typedef enum {
   MP_LT = -1,    /* less than */
   MP_EQ = 0,     /* equal */
   MP_GT = 1      /* greater than */
} mp_ord;
.in -1.5i
.in -1i

.LP
.BI mp_err
.in 1i
The handful of different errors Libtommath can throw.

.in 1.5i
.nf
typedef enum {
   MP_OKAY  = 0,   /* no error */
   MP_ERR   = -1,  /* unknown error */
   MP_MEM   = -2,  /* out of mem */
   MP_VAL   = -3,  /* invalid input */
   MP_ITER  = -4,  /* maximum iterations reached */
   MP_BUF   = -5,  /* buffer overflow, supplied buffer too small */
   MP_OVF   = -6   /* mp_int overflow, too many digits */
} mp_err;
.in -1.5i
.in -1i

.LP
.BI mp_order
.in 1i
Bit order. If the most signifcant bit comes first or the least significant one.
.br
Not to be mistaken for \fBmp_ord\fP.

.in 1.5i
.nf
typedef enum {
   MP_LSB_FIRST = -1,
   MP_MSB_FIRST =  1
} mp_order;
.in -1.5i
.in -1i

.LP
.BI mp_endian
.in 1i
Byte order. Only big and little endian are supported. \fBMP_NATIVE_ENDIAN\fP refers to the local (detected)
endianess.

.in 1.5i
.nf
typedef enum {
   MP_LITTLE_ENDIAN  = -1,
   MP_NATIVE_ENDIAN  =  0,
   MP_BIG_ENDIAN     =  1
} mp_endian;
.in -1.5i
.in -1i

.LP
.BI hval
.in 1i
Type holding the hash produced by \fBmp_hash\fP. Either a 32 bit or a 64 bit unsigned type.
.in -1i

.LP
.SS "MACROS"
.LP
.BI MP_DIGIT_BIT
.in 1i
The size in bits of a limb.
.br
The value depends on the size of the biggest available native integer type.
.TS
tab(;) allbox;
c c c
c l l.
\fBarch bits\fP;\fBMP_DIGIT_BIT\fP;\fBMACRO\fP
16;15;MP_16BIT
32;28, (31);MP_32BIT, (MP_31BIT) (1)
64;60, (28);MP_64BIT, (MP_64BIT) (2)
32/64;28;MP_28BIT (3)
.TE

(1) The size 31 bit reduces some of the optimizations, especially COMBA.
.br
(2) Some C-compilers do not offer 16 byte integers. One of them is very famous.
.br
(3) Default when all tests to search for something better fail.

.br
\fBNOTE:\fP ISO C 23 introduced \fB_BitInt(n)\fP which would allow 16 byte integers even
on non 64 bit architectures. Not much support as of the time of this writing (LLVM does, for example).
.in -1i

.LP
.BI MP_PRIME_BBS
.in 1i
Generate a BBS style prime. (See: mp_prime_rand)
.in -1i

.LP
.BI MP_PRIME_SAFE
.in 1i
Generate a safe prime such that (p-1)/2 == prime. (See: mp_prime_rand)
.in -1i

.LP
.BI MP_PRIME_2MSB_ON
.in 1i
Force 2nd most siginficant bit to be 1 (one). when generating primes. (See: mp_prime_rand)
.in -1i

.LP
.BI MP_DIGIT_MAX
.in 1i
Largest value a \fBmp_digit\fP can hold: \fB-1 + 2^MP_DIGIT_BIT\fP.
.in -1i

.LP
.BI MP_MASK
.in 1i
Holds the same value as \fBMP_DIGIT_MAX\fP.
.br
\fBNOTE:\fP this macro computes the value, \fBMP_DIGIT_MAX\fP is a copy.
.in -1i


.LP
.SS "GLOBAL VARIABLES"
.\" TODO: change if COMBA gets its heap alternative
Libtommath uses no global variables, it is threadsafe.
.br
But some global variables are there to allow for e.g.: tuning and can hence be abused. But shouldn't.

.LP
.BI "extern int MP_MUL_KARATSUBA_CUTOFF"
.in 1i
Sets the cutoff value when Karatsuba multiplication comes in if the macro \fBMP_FIXED_CUTOFFS\fP is not defined.
.in -1i


.LP
.BI "extern int MP_SQR_KARATSUBA_CUTOFF"
.in 1i
Sets the cutoff value when Karatsuba squaring comes in if the macro \fBMP_FIXED_CUTOFFS\fP is not defined.
.in -1i


.LP
.BI "extern int MP_MUL_TOOM_CUTOFF"
.in 1i
Sets the cutoff value when Toom-Cook 3-way multiplication comes in if the macro \fBMP_FIXED_CUTOFFS\fP is not defined.
.in -1i


.LP
.BI "extern int MP_SQR_TOOM_CUTOFF"
.in 1i
Sets the cutoff value when Toom-Cook 3-way squaring comes in if the macro \fBMP_FIXED_CUTOFFS\fP is not defined.
.in -1i

.LP
.SH "EXAMPLES"
.\" TODO: does it even make sense to have examples?
.LP
.SS "Is \fIn\fP a Perfect Power"
A function that uses a mix of libtommath's functions.

.nf
static mp_err mp_is_perfect_power(const mp_int *n, bool *result,
                                    mp_int *rootout, mp_int *exponent)
{
   mp_int root, power, prime, max;
   int highbit, lowbit, p;
   mp_err err = MP_OKAY;

   *result = false;

   /* No negative numbers. For now. */
   if (mp_cmp_d(n, 4) == MP_LT) {
      err = MP_VAL;
      goto LTM_OUT;
   }

   /* Compute floor(log_2(n)) */
   highbit = mp_count_bits(n) - 1;

   /* MP_IS_POWER_OF_TWO(n) is a macro in tommath_private.h */
   if (MP_IS_POWER_OF_TWO(n)) {
      *result = true;
      if (exponent != NULL) {
         if ((err = mp_set_l(exponent, (long)highbit)) != MP_OKAY) {
            return err;
         }
      }
      if (rootout != NULL) {
         if ((err = mp_set_l(rootout, 2l)) != MP_OKAY) {
            return err;
         }
      }
      return err;
   }
   /* Initialize the needed variables all at once */
   if ((err = mp_init_multi(&root, &power, &prime, &max, NULL)) != MP_OKAY) {
      return err;
   }
   /* a long is at least as big as an int */
   if( (err = mp_set_l(&max, (long)highbit) ) != MP_OKAY)                goto LTM_ERR;

   /* mp_prime_next_prime() returns the next prime, so start with a preceding value */
   mp_set(&prime, 2u);

   while (mp_cmp(&prime, &max) != MP_GT) {
      /* Deterministic up to 2^82 but the largest value possible is < INT_MAX       */
      if ((err = mp_prime_next_prime(&prime, -1, false)) != MP_OKAY)     goto LTM_ERR;
      p = (int)mp_get_l(&prime);
      if ((err = mp_root_n(n, p, &root)) != MP_OKAY)                     goto LTM_ERR;
      if ((err = mp_expt_n(&root, p, &power)) != MP_OKAY)                goto LTM_ERR;

      if (mp_cmp(n, &power) == MP_EQ) {
         *result = true;
         if (rootout != NULL) {
            mp_exch(&root, rootout);
         }
         if (exponent != NULL) {
            if ((e = mp_set_l(exponent, (long)p)) != MP_OKAY)            goto LTM_ERR;
         }
      }
   }

LTM_OUT:
   if (rootout != NULL) {
      if ((err = mp_set(rootout, 0u)) != MP_OKAY)                        goto LTM_ERR;
   }
   if (exponent != NULL) {
      if ((err = mp_set(exponent, 0u)) != MP_OKAY)                       goto LTM_ERR;
   }
LTM_ERR:
   mp_clear_multi(&root, &power,&prime,  NULL);
   return err;
}
.LP
.SS "Pi by Binary Splitting"
Compute digits of pi using Manchin's formula from 1706. A classic.

.nf
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <string.h>
#include <limits.h>
#include <tommath.h>

static int mp_acot_binary_splitting(
             mp_int * q, mp_int * a, mp_int * b, mp_int * P,
             mp_int * Q, mp_int * R, int * idx)
{
    mp_err err;
    mp_int p1, q1, r1, p2, q2, r2, t1, t2, one;
    if ((err = mp_init_multi(
                  &p1, &q1, &r1, &p2, &q2,
                  &r2, &t1, &t2, &one, NULL)) != MP_OKAY) {
        return err;
    }

    err = MP_OKAY;
    mp_set(&one, 1);

    if ((err = mp_sub(b, a, &t1)) != MP_OKAY)                 goto LTM_ERR;
    if (mp_cmp(&t1, &one) == MP_EQ) {
        if ((err = mp_mul_2d(a, 1, &t1)) != MP_OKAY)          goto LTM_ERR;
        if ((err = mp_add_d(&t1, 3, &t1)) != MP_OKAY)         goto LTM_ERR;
        mp_set(P, 1);
        if ( ((*idx) & 1) == 0 ) {
            P->sign = MP_NEG;
        }
        (*idx)++;
        if ((err = mp_mul(&t1, q, Q)) != MP_OKAY)             goto LTM_ERR;
        if ((err = mp_copy(&t1, R)) != MP_OKAY)               goto LTM_ERR;
        /* Done */
        goto LTM_ERR;
    }

    if ((err = mp_add(a, b, &t1)) != MP_OKAY)                 goto LTM_ERR;
    if ((err = mp_div_2d(&t1, 1, &t1, NULL)) != MP_OKAY)      goto LTM_ERR;


    if ((err = mp_acot_binary_splitting(
                  q, a, &t1, &p1, &q1, &r1, idx)) != MP_OKAY) goto LTM_ERR;
    if ((err = mp_acot_binary_splitting(
                  q, &t1, b, &p2, &q2, &r2, idx)) != MP_OKAY) goto LTM_ERR;

    /* P = q2*p1 + r1*p2 */
    if ((err = mp_mul(&q2, &p1, &t1)) != MP_OKAY)             goto LTM_ERR;
    if ((err = mp_mul(&r1, &p2, &t2)) != MP_OKAY)             goto LTM_ERR;
    if ((err = mp_add(&t1, &t2, P)) != MP_OKAY)               goto LTM_ERR;
    /* Q =  q1*q2 */
    if ((err = mp_mul(&q1, &q2, Q)) != MP_OKAY)               goto LTM_ERR;
    /* R = r1*r2 */
    if ((err = mp_mul(&r1, &r2, R)) != MP_OKAY)               goto LTM_ERR;

  LTM_ERR:
    mp_clear_multi(&p1, &q1, &r1, &p2, &q2, &r2, &t1, &t2, &one, NULL);
    return err;
}


int main(int argc, char **argv)
{
   mp_int N, P, Q, R, zero, EPS, t1, t2, t5, t239;
   mp_err err = MP_OKAY;
   int idx = 0;
   long eps, dec;

   if(argc != 2) {
     fprintf(stderr, "Usage %s bits of precision", argv[0]);
     exit(EXIT_FAILURE);
   }

   errno = 0;
   eps = strtol(argv[1],NULL,10);
   if ((errno == ERANGE && (eps == LONG_MAX || eps == LONG_MIN))
       || (errno != 0 && eps == 0)) {
      fprintf(stderr,"Error reading precision. Reason: %s\n",
                     strerror(errno));
      exit(EXIT_FAILURE);
   }

   if( (err = mp_init_multi(
       &N, &P, &Q, &R, &zero, &EPS, &t1, &t2, &t5, &t239, NULL) ) != MP_OKAY) {
      fprintf(stderr,"Error initializing mp_ints. Reason: %s\n",
                     mp_error_to_string(err));
      exit(EXIT_FAILURE);
   }

   /* Machin's 1706 */
   mp_set(&t5, 5u);
   mp_set(&t239, 239u);
   mp_set_l(&EPS, eps);

   if( (err = mp_sqr(&t5, &t1) ) != MP_OKAY)                      goto LTM_ERR;
   if( (err = mp_acot_binary_splitting(
             &t1, &zero, &EPS, &P, &Q, &R, &idx) ) != MP_OKAY)    goto LTM_ERR;
   if( (err = mp_add(&P, &Q, &t1) ) != MP_OKAY)                   goto LTM_ERR;
   if( (err = mp_mul_2d(&t1, (int)eps, &t1) ) != MP_OKAY)         goto LTM_ERR;
   if( (err = mp_mul(&Q, &t5, &t2) ) != MP_OKAY)                  goto LTM_ERR;
   if( (err = mp_div(&t1, &t2, &t5, NULL) ) != MP_OKAY)           goto LTM_ERR;

   if( (err = mp_sqr(&t239, &t1) ) != MP_OKAY)                    goto LTM_ERR;
   if( (err = mp_acot_binary_splitting(
             &t1, &zero, &EPS, &P, &Q, &R, &idx) ) != MP_OKAY)    goto LTM_ERR;
   if( (err = mp_add(&P, &Q, &t1) ) != MP_OKAY)                   goto LTM_ERR;
   if( (err = mp_mul_2d(&t1, (int)eps, &t1) ) != MP_OKAY)         goto LTM_ERR;
   if( (err = mp_mul(&Q, &t239, &t2) ) != MP_OKAY)                goto LTM_ERR;
   if( (err = mp_div(&t1, &t2, &t239, NULL) ) != MP_OKAY)         goto LTM_ERR;

   if( (err = mp_mul_2d(&t5, 2, &t5) ) != MP_OKAY)                goto LTM_ERR;
   if( (err = mp_sub(&t5, &t239, &t1) ) != MP_OKAY)               goto LTM_ERR;
   if( (err = mp_mul_2d(&t1, 2, &t1) ) != MP_OKAY)                goto LTM_ERR;

   /* 146/485 ~ 1/log_2(10) for about 7 decimal digits */
   dec = (eps * 146) / 485;
   mp_set(&t2, 10);
   if( (err = mp_expt_n(&t2, (int)dec, &t2) ) != MP_OKAY)         goto LTM_ERR;
   if( (err = mp_mul(&t1,&t2,&t1) ) != MP_OKAY)                   goto LTM_ERR;
   if( (err = mp_div_2d(&t1, (int)eps, &t1, NULL) ) != MP_OKAY)   goto LTM_ERR;

   if( (err = mp_fwrite(&t1,10,stdout) ) != MP_OKAY)              goto LTM_ERR;

   mp_clear_multi(&N, &P, &Q, &R, &zero, &EPS, &t1, &t2, &t5, &t239, NULL);
   exit(EXIT_SUCCESS);
LTM_ERR:
   mp_clear_multi(&N, &P, &Q, &R, &zero, &EPS, &t1, &t2, &t5, &t239, NULL);
   exit(EXIT_FAILURE);
}



.LP
.SH "SEE ALSO"

.IR https://github.com/libtom/libtommath

.SH BUGS

Please report all bugs and other incommodities at
.IR https://github.com/libtom/libtommath/issues

.SH AUTHORS AND LICENSE
Version 1.2.0
.LP
Copyright (C) 2002 Tom St Denis

.ad c

The LibTom license

.ad n
.LP
This is free and unencumbered software released into the public domain.
.LP
Anyone is free to copy, modify, publish, use, compile, sell, or
distribute this software, either in source code form or as a compiled
binary, for any purpose, commercial or non-commercial, and by any
means.
.LP
In jurisdictions that recognize copyright laws, the author or authors
of this software dedicate any and all copyright interest in the
software to the public domain. We make this dedication for the benefit
of the public at large and to the detriment of our heirs and
successors. We intend this dedication to be an overt act of
relinquishment in perpetuity of all present and future rights to this
software under copyright law.
.LP
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
OTHER DEALINGS IN THE SOFTWARE.
.LP
For more information, please refer to <http://unlicense.org/>

." end of man page