void mp_binary_set_val(char struct_type, char val_type, mp_obj_t val_in, byte **ptr) {
byte *p = *ptr;
mp_uint_t align;
size_t size = mp_binary_get_size(struct_type, val_type, &align);
if (struct_type == '@') {
// Make pointer aligned
p = (byte*)MP_ALIGN(p, (size_t)align);
if (MP_ENDIANNESS_LITTLE) {
struct_type = '<';
} else {
struct_type = '>';
}
}
*ptr = p + size;
mp_uint_t val;
switch (val_type) {
case 'O':
val = (mp_uint_t)val_in;
break;
#if MICROPY_PY_BUILTINS_FLOAT
case 'f': {
union { uint32_t i; float f; } fp_sp;
fp_sp.f = mp_obj_get_float(val_in);
val = fp_sp.i;
break;
}
case 'd': {
union { uint64_t i64; uint32_t i32[2]; double f; } fp_dp;
fp_dp.f = mp_obj_get_float(val_in);
if (BYTES_PER_WORD == 8) {
val = fp_dp.i64;
} else {
int be = struct_type == '>';
mp_binary_set_int(sizeof(uint32_t), be, p, fp_dp.i32[MP_ENDIANNESS_BIG ^ be]);
p += sizeof(uint32_t);
val = fp_dp.i32[MP_ENDIANNESS_LITTLE ^ be];
}
break;
}
#endif
default:
#if MICROPY_LONGINT_IMPL != MICROPY_LONGINT_IMPL_NONE
if (MP_OBJ_IS_TYPE(val_in, &mp_type_int)) {
mp_obj_int_to_bytes_impl(val_in, struct_type == '>', size, p);
return;
} else
#endif
{
val = mp_obj_get_int(val_in);
// sign extend if needed
if (BYTES_PER_WORD < 8 && size > sizeof(val) && is_signed(val_type) && (mp_int_t)val < 0) {
memset(p + sizeof(val), 0xff, size - sizeof(val));
}
}
}
mp_binary_set_int(MIN((size_t)size, sizeof(val)), struct_type == '>', p, val);
}
mp_obj_t mp_binary_get_val(char struct_type, char val_type, byte **ptr) {
byte *p = *ptr;
mp_uint_t align;
size_t size = mp_binary_get_size(struct_type, val_type, &align);
if (struct_type == '@') {
// Make pointer aligned
p = (byte*)MP_ALIGN(p, (size_t)align);
#if MP_ENDIANNESS_LITTLE
struct_type = '<';
#else
struct_type = '>';
#endif
}
*ptr = p + size;
long long val = mp_binary_get_int(size, is_signed(val_type), (struct_type == '>'), p);
if (val_type == 'O') {
return (mp_obj_t)(mp_uint_t)val;
} else if (val_type == 'S') {
const char *s_val = (const char*)(uintptr_t)(mp_uint_t)val;
return mp_obj_new_str(s_val, strlen(s_val), false);
#if MICROPY_PY_BUILTINS_FLOAT
} else if (val_type == 'f') {
union { uint32_t i; float f; } fpu = {val};
return mp_obj_new_float(fpu.f);
} else if (val_type == 'd') {
union { uint64_t i; double f; } fpu = {val};
return mp_obj_new_float(fpu.f);
#endif
} else if (is_signed(val_type)) {
if ((long long)MP_SMALL_INT_MIN <= val && val <= (long long)MP_SMALL_INT_MAX) {
return mp_obj_new_int((mp_int_t)val);
} else {
return mp_obj_new_int_from_ll(val);
}
} else {
if ((unsigned long long)val <= (unsigned long long)MP_SMALL_INT_MAX) {
return mp_obj_new_int_from_uint((mp_uint_t)val);
} else {
return mp_obj_new_int_from_ull(val);
}
}
}
/* Do modular exponentiation using integer multiply code. */
mp_err mp_exptmod_safe_i(const mp_int * montBase,
const mp_int * exponent,
const mp_int * modulus,
mp_int * result,
mp_mont_modulus *mmm,
int nLen,
mp_size bits_in_exponent,
mp_size window_bits,
mp_size num_powers)
{
mp_int *pa1, *pa2, *ptmp;
mp_size i;
mp_size first_window;
mp_err res;
int expOff;
mp_int accum1, accum2, accum[WEAVE_WORD_SIZE];
mp_int tmp;
unsigned char *powersArray;
unsigned char *powers;
MP_DIGITS(&accum1) = 0;
MP_DIGITS(&accum2) = 0;
MP_DIGITS(&accum[0]) = 0;
MP_DIGITS(&accum[1]) = 0;
MP_DIGITS(&accum[2]) = 0;
MP_DIGITS(&accum[3]) = 0;
MP_DIGITS(&tmp) = 0;
powersArray = (unsigned char *)malloc(num_powers*(nLen*sizeof(mp_digit)+1));
if (powersArray == NULL) {
res = MP_MEM;
goto CLEANUP;
}
/* powers[i] = base ** (i); */
powers = (unsigned char *)MP_ALIGN(powersArray,num_powers);
/* grab the first window value. This allows us to preload accumulator1
* and save a conversion, some squares and a multiple*/
MP_CHECKOK( mpl_get_bits(exponent,
bits_in_exponent-window_bits, window_bits) );
first_window = (mp_size)res;
MP_CHECKOK( mp_init_size(&accum1, 3 * nLen + 2) );
MP_CHECKOK( mp_init_size(&accum2, 3 * nLen + 2) );
MP_CHECKOK( mp_init_size(&tmp, 3 * nLen + 2) );
/* build the first WEAVE_WORD powers inline */
/* if WEAVE_WORD_SIZE is not 4, this code will have to change */
if (num_powers > 2) {
MP_CHECKOK( mp_init_size(&accum[0], 3 * nLen + 2) );
MP_CHECKOK( mp_init_size(&accum[1], 3 * nLen + 2) );
MP_CHECKOK( mp_init_size(&accum[2], 3 * nLen + 2) );
MP_CHECKOK( mp_init_size(&accum[3], 3 * nLen + 2) );
mp_set(&accum[0], 1);
MP_CHECKOK( s_mp_to_mont(&accum[0], mmm, &accum[0]) );
MP_CHECKOK( mp_copy(montBase, &accum[1]) );
SQR(montBase, &accum[2]);
MUL_NOWEAVE(montBase, &accum[2], &accum[3]);
MP_CHECKOK( mpi_to_weave(accum, powers, nLen, num_powers) );
if (first_window < 4) {
MP_CHECKOK( mp_copy(&accum[first_window], &accum1) );
first_window = num_powers;
}
} else {
if (first_window == 0) {
mp_set(&accum1, 1);
MP_CHECKOK( s_mp_to_mont(&accum1, mmm, &accum1) );
} else {
/* assert first_window == 1? */
MP_CHECKOK( mp_copy(montBase, &accum1) );
}
}
/*
* calculate all the powers in the powers array.
* this adds 2**(k-1)-2 square operations over just calculating the
* odd powers where k is the window size in the two other mp_modexpt
* implementations in this file. We will get some of that
* back by not needing the first 'k' squares and one multiply for the
* first window */
for (i = WEAVE_WORD_SIZE; i < num_powers; i++) {
int acc_index = i & (WEAVE_WORD_SIZE-1); /* i % WEAVE_WORD_SIZE */
if ( i & 1 ) {
MUL_NOWEAVE(montBase, &accum[acc_index-1] , &accum[acc_index]);
/* we've filled the array do our 'per array' processing */
if (acc_index == (WEAVE_WORD_SIZE-1)) {
MP_CHECKOK( mpi_to_weave(accum, powers + i - (WEAVE_WORD_SIZE-1),
nLen, num_powers) );
if (first_window <= i) {
MP_CHECKOK( mp_copy(&accum[first_window & (WEAVE_WORD_SIZE-1)],
&accum1) );
first_window = num_powers;
}
}
} else {
/* up to 8 we can find 2^i-1 in the accum array, but at 8 we our source
* and target are the same so we need to copy.. After that, the
* value is overwritten, so we need to fetch it from the stored
* weave array */
if (i > 2* WEAVE_WORD_SIZE) {
MP_CHECKOK(weave_to_mpi(&accum2, powers+i/2, nLen, num_powers));
SQR(&accum2, &accum[acc_index]);
} else {
int half_power_index = (i/2) & (WEAVE_WORD_SIZE-1);
if (half_power_index == acc_index) {
/* copy is cheaper than weave_to_mpi */
MP_CHECKOK(mp_copy(&accum[half_power_index], &accum2));
SQR(&accum2,&accum[acc_index]);
} else {
SQR(&accum[half_power_index],&accum[acc_index]);
}
}
}
}
/* if the accum1 isn't set, Then there is something wrong with our logic
* above and is an internal programming error.
*/
#if MP_ARGCHK == 2
assert(MP_USED(&accum1) != 0);
#endif
/* set accumulator to montgomery residue of 1 */
pa1 = &accum1;
pa2 = &accum2;
for (expOff = bits_in_exponent - window_bits*2; expOff >= 0; expOff -= window_bits) {
mp_size smallExp;
MP_CHECKOK( mpl_get_bits(exponent, expOff, window_bits) );
smallExp = (mp_size)res;
/* handle unroll the loops */
switch (window_bits) {
case 1:
if (!smallExp) {
SQR(pa1,pa2); SWAPPA;
} else if (smallExp & 1) {
SQR(pa1,pa2); MUL_NOWEAVE(montBase,pa2,pa1);
} else {
abort();
}
break;
case 6:
SQR(pa1,pa2); SQR(pa2,pa1);
/* fall through */
case 4:
SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1);
MUL(smallExp, pa1,pa2); SWAPPA;
break;
case 5:
SQR(pa1,pa2); SQR(pa2,pa1); SQR(pa1,pa2); SQR(pa2,pa1);
SQR(pa1,pa2); MUL(smallExp,pa2,pa1);
break;
default:
abort(); /* could do a loop? */
}
}
res = s_mp_redc(pa1, mmm);
mp_exch(pa1, result);
CLEANUP:
mp_clear(&accum1);
mp_clear(&accum2);
mp_clear(&accum[0]);
mp_clear(&accum[1]);
mp_clear(&accum[2]);
mp_clear(&accum[3]);
mp_clear(&tmp);
/* PORT_Memset(powers,0,num_powers*nLen*sizeof(mp_digit)); */
free(powersArray);
return res;
}
