trim trailing spaces

This commit is contained in:
Steffen Jaeckel 2014-05-09 23:00:00 +02:00
parent dc31ca545f
commit 5d2fe0da8c
3 changed files with 34 additions and 34 deletions

View File

@ -18,7 +18,7 @@
#include <gmp.h>
static int init(void **a)
{
{
LTC_ARGCHK(a != NULL);
*a = XCALLOC(1, sizeof(__mpz_struct));
@ -85,7 +85,7 @@ static int get_digit_count(void *a)
LTC_ARGCHK(a != NULL);
return mpz_size(a);
}
static int compare(void *a, void *b)
{
int ret;
@ -234,7 +234,7 @@ static int add(void *a, void *b, void *c)
mpz_add(c, a, b);
return CRYPT_OK;
}
static int addi(void *a, unsigned long b, void *c)
{
LTC_ARGCHK(a != NULL);
@ -321,10 +321,10 @@ static int modi(void *a, unsigned long b, unsigned long *c)
{
LTC_ARGCHK(a != NULL);
LTC_ARGCHK(c != NULL);
*c = mpz_fdiv_ui(a, b);
return CRYPT_OK;
}
}
/* gcd */
static int gcd(void *a, void *b, void *c)
@ -440,7 +440,7 @@ static int exptmod(void *a, void *b, void *c, void *d)
LTC_ARGCHK(d != NULL);
mpz_powm(d, a, b, c);
return CRYPT_OK;
}
}
static int isprime(void *a, int *b)
{
@ -539,7 +539,7 @@ const ltc_math_descriptor gmp_desc = {
#endif
&addmod,
&submod,
&set_rand,
};

View File

@ -25,7 +25,7 @@ static const struct {
};
/**
Convert a MPI error to a LTC error (Possibly the most powerful function ever! Oh wait... no)
Convert a MPI error to a LTC error (Possibly the most powerful function ever! Oh wait... no)
@param err The error to convert
@return The equivalent LTC error code or CRYPT_ERROR if none found
*/
@ -34,7 +34,7 @@ static int mpi_to_ltc_error(int err)
int x;
for (x = 0; x < (int)(sizeof(mpi_to_ltc_codes)/sizeof(mpi_to_ltc_codes[0])); x++) {
if (err == mpi_to_ltc_codes[x].mpi_code) {
if (err == mpi_to_ltc_codes[x].mpi_code) {
return mpi_to_ltc_codes[x].ltc_code;
}
}
@ -51,7 +51,7 @@ static int init(void **a)
if (*a == NULL) {
return CRYPT_MEM;
}
if ((err = mpi_to_ltc_error(mp_init(*a))) != CRYPT_OK) {
XFREE(*a);
}
@ -115,7 +115,7 @@ static int get_digit_count(void *a)
A = a;
return A->used;
}
static int compare(void *a, void *b)
{
int ret;
@ -211,7 +211,7 @@ static int add(void *a, void *b, void *c)
LTC_ARGCHK(c != NULL);
return mpi_to_ltc_error(mp_add(a, b, c));
}
static int addi(void *a, unsigned long b, void *c)
{
LTC_ARGCHK(a != NULL);
@ -288,7 +288,7 @@ static int modi(void *a, unsigned long b, unsigned long *c)
}
*c = tmp;
return CRYPT_OK;
}
}
/* gcd */
static int gcd(void *a, void *b, void *c)
@ -398,7 +398,7 @@ static int exptmod(void *a, void *b, void *c, void *d)
LTC_ARGCHK(c != NULL);
LTC_ARGCHK(d != NULL);
return mpi_to_ltc_error(mp_exptmod(a,b,c,d));
}
}
static int isprime(void *a, int *b)
{
@ -460,7 +460,7 @@ const ltc_math_descriptor ltm_desc = {
&mulmod,
&sqrmod,
&invmod,
&montgomery_setup,
&montgomery_normalization,
&montgomery_reduce,
@ -472,7 +472,7 @@ const ltc_math_descriptor ltm_desc = {
#ifdef LTC_MECC
#ifdef LTC_MECC_FP
&ltc_ecc_fp_mulmod,
#else
#else
&ltc_ecc_mulmod,
#endif
&ltc_ecc_projective_add_point,
@ -499,7 +499,7 @@ const ltc_math_descriptor ltm_desc = {
#endif
&addmod,
&submod,
&set_rand,
};

View File

@ -25,7 +25,7 @@ static const struct {
};
/**
Convert a tfm error to a LTC error (Possibly the most powerful function ever! Oh wait... no)
Convert a tfm error to a LTC error (Possibly the most powerful function ever! Oh wait... no)
@param err The error to convert
@return The equivalent LTC error code or CRYPT_ERROR if none found
*/
@ -34,7 +34,7 @@ static int tfm_to_ltc_error(int err)
int x;
for (x = 0; x < (int)(sizeof(tfm_to_ltc_codes)/sizeof(tfm_to_ltc_codes[0])); x++) {
if (err == tfm_to_ltc_codes[x].tfm_code) {
if (err == tfm_to_ltc_codes[x].tfm_code) {
return tfm_to_ltc_codes[x].ltc_code;
}
}
@ -114,7 +114,7 @@ static int get_digit_count(void *a)
A = a;
return A->used;
}
static int compare(void *a, void *b)
{
int ret;
@ -213,7 +213,7 @@ static int add(void *a, void *b, void *c)
fp_add(a, b, c);
return CRYPT_OK;
}
static int addi(void *a, unsigned long b, void *c)
{
LTC_ARGCHK(a != NULL);
@ -246,7 +246,7 @@ static int mul(void *a, void *b, void *c)
LTC_ARGCHK(a != NULL);
LTC_ARGCHK(b != NULL);
LTC_ARGCHK(c != NULL);
fp_mul(a, b, c);
fp_mul(a, b, c);
return CRYPT_OK;
}
@ -297,7 +297,7 @@ static int modi(void *a, unsigned long b, unsigned long *c)
}
*c = tmp;
return CRYPT_OK;
}
}
/* gcd */
static int gcd(void *a, void *b, void *c)
@ -411,7 +411,7 @@ static int exptmod(void *a, void *b, void *c, void *d)
LTC_ARGCHK(c != NULL);
LTC_ARGCHK(d != NULL);
return tfm_to_ltc_error(fp_exptmod(a,b,c,d));
}
}
static int isprime(void *a, int *b)
{
@ -455,7 +455,7 @@ static int tfm_ecc_projective_dbl_point(ecc_point *P, ecc_point *R, void *modulu
if (fp_cmp(R->z, modulus) != FP_LT) {
fp_sub(R->z, modulus, R->z);
}
/* &t2 = X - T1 */
fp_sub(R->x, &t1, &t2);
if (fp_cmp_d(&t2, 0) == FP_LT) {
@ -514,7 +514,7 @@ static int tfm_ecc_projective_dbl_point(ecc_point *P, ecc_point *R, void *modulu
fp_add(R->x, modulus, R->x);
}
/* Y = Y - X */
/* Y = Y - X */
fp_sub(R->y, R->x, R->y);
if (fp_cmp_d(R->y, 0) == FP_LT) {
fp_add(R->y, modulus, R->y);
@ -527,7 +527,7 @@ static int tfm_ecc_projective_dbl_point(ecc_point *P, ecc_point *R, void *modulu
if (fp_cmp_d(R->y, 0) == FP_LT) {
fp_add(R->y, modulus, R->y);
}
return CRYPT_OK;
}
@ -543,8 +543,8 @@ static int tfm_ecc_projective_dbl_point(ecc_point *P, ecc_point *R, void *modulu
static int tfm_ecc_projective_add_point(ecc_point *P, ecc_point *Q, ecc_point *R, void *modulus, void *Mp)
{
fp_int t1, t2, x, y, z;
fp_digit mp;
fp_digit mp;
LTC_ARGCHK(P != NULL);
LTC_ARGCHK(Q != NULL);
LTC_ARGCHK(R != NULL);
@ -561,7 +561,7 @@ static int tfm_ecc_projective_add_point(ecc_point *P, ecc_point *Q, ecc_point *R
/* should we dbl instead? */
fp_sub(modulus, Q->y, &t1);
if ( (fp_cmp(P->x, Q->x) == FP_EQ) &&
if ( (fp_cmp(P->x, Q->x) == FP_EQ) &&
(Q->z != NULL && fp_cmp(P->z, Q->z) == FP_EQ) &&
(fp_cmp(P->y, Q->y) == FP_EQ || fp_cmp(P->y, &t1) == FP_EQ)) {
return tfm_ecc_projective_dbl_point(P, R, modulus, Mp);
@ -654,7 +654,7 @@ static int tfm_ecc_projective_add_point(ecc_point *P, ecc_point *Q, ecc_point *R
/* T1 = T1 * X */
fp_mul(&t1, &x, &t1);
fp_montgomery_reduce(&t1, modulus, mp);
/* X = Y*Y */
fp_sqr(&y, &x);
fp_montgomery_reduce(&x, modulus, mp);
@ -668,7 +668,7 @@ static int tfm_ecc_projective_add_point(ecc_point *P, ecc_point *Q, ecc_point *R
fp_sub(&t2, &x, &t2);
if (fp_cmp_d(&t2, 0) == FP_LT) {
fp_add(&t2, modulus, &t2);
}
}
/* T2 = T2 - X */
fp_sub(&t2, &x, &t2);
if (fp_cmp_d(&t2, 0) == FP_LT) {
@ -691,7 +691,7 @@ static int tfm_ecc_projective_add_point(ecc_point *P, ecc_point *Q, ecc_point *R
fp_copy(&x, R->x);
fp_copy(&y, R->y);
fp_copy(&z, R->z);
return CRYPT_OK;
}
@ -786,7 +786,7 @@ const ltc_math_descriptor tfm_desc = {
#endif
&addmod,
&submod,
NULL,
};