1 /* mpn_mod_1(dividend_ptr, dividend_size, divisor_limb) --
2 Divide (DIVIDEND_PTR,,DIVIDEND_SIZE) by DIVISOR_LIMB.
3 Return the single-limb remainder.
4 There are no constraints on the value of the divisor.
6 Copyright 1991, 1993, 1994, 1999, 2000, 2002, 2007-2009, 2012 Free Software
9 This file is part of the GNU MP Library.
11 The GNU MP Library is free software; you can redistribute it and/or modify
12 it under the terms of either:
14 * the GNU Lesser General Public License as published by the Free
15 Software Foundation; either version 3 of the License, or (at your
16 option) any later version.
20 * the GNU General Public License as published by the Free Software
21 Foundation; either version 2 of the License, or (at your option) any
24 or both in parallel, as here.
26 The GNU MP Library is distributed in the hope that it will be useful, but
27 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
28 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
31 You should have received copies of the GNU General Public License and the
32 GNU Lesser General Public License along with the GNU MP Library. If not,
33 see https://www.gnu.org/licenses/. */
40 /* The size where udiv_qrnnd_preinv should be used rather than udiv_qrnnd,
41 meaning the quotient size where that should happen, the quotient size
42 being how many udiv divisions will be done.
44 The default is to use preinv always, CPUs where this doesn't suit have
45 tuned thresholds. Note in particular that preinv should certainly be
46 used if that's the only division available (USE_PREINV_ALWAYS). */
48 #ifndef MOD_1_NORM_THRESHOLD
49 #define MOD_1_NORM_THRESHOLD 0
52 #ifndef MOD_1_UNNORM_THRESHOLD
53 #define MOD_1_UNNORM_THRESHOLD 0
56 #ifndef MOD_1U_TO_MOD_1_1_THRESHOLD
57 #define MOD_1U_TO_MOD_1_1_THRESHOLD MP_SIZE_T_MAX /* default is not to use mpn_mod_1s */
60 #ifndef MOD_1N_TO_MOD_1_1_THRESHOLD
61 #define MOD_1N_TO_MOD_1_1_THRESHOLD MP_SIZE_T_MAX /* default is not to use mpn_mod_1s */
64 #ifndef MOD_1_1_TO_MOD_1_2_THRESHOLD
65 #define MOD_1_1_TO_MOD_1_2_THRESHOLD 10
68 #ifndef MOD_1_2_TO_MOD_1_4_THRESHOLD
69 #define MOD_1_2_TO_MOD_1_4_THRESHOLD 20
72 #if TUNE_PROGRAM_BUILD && !HAVE_NATIVE_mpn_mod_1_1p
73 /* Duplicates declarations in tune/speed.h */
74 mp_limb_t
mpn_mod_1_1p_1 (mp_srcptr
, mp_size_t
, mp_limb_t
, mp_limb_t
[4]);
75 mp_limb_t
mpn_mod_1_1p_2 (mp_srcptr
, mp_size_t
, mp_limb_t
, mp_limb_t
[4]);
77 void mpn_mod_1_1p_cps_1 (mp_limb_t
[4], mp_limb_t
);
78 void mpn_mod_1_1p_cps_2 (mp_limb_t
[4], mp_limb_t
);
81 #define mpn_mod_1_1p(ap, n, b, pre) \
82 (mod_1_1p_method == 1 ? mpn_mod_1_1p_1 (ap, n, b, pre) \
83 : (mod_1_1p_method == 2 ? mpn_mod_1_1p_2 (ap, n, b, pre) \
84 : __gmpn_mod_1_1p (ap, n, b, pre)))
86 #undef mpn_mod_1_1p_cps
87 #define mpn_mod_1_1p_cps(pre, b) \
88 (mod_1_1p_method == 1 ? mpn_mod_1_1p_cps_1 (pre, b) \
89 : (mod_1_1p_method == 2 ? mpn_mod_1_1p_cps_2 (pre, b) \
90 : __gmpn_mod_1_1p_cps (pre, b)))
91 #endif /* TUNE_PROGRAM_BUILD && !HAVE_NATIVE_mpn_mod_1_1p */
94 /* The comments in mpn/generic/divrem_1.c apply here too.
96 As noted in the algorithms section of the manual, the shifts in the loop
97 for the unnorm case can be avoided by calculating r = a%(d*2^n), followed
98 by a final (r*2^n)%(d*2^n). In fact if it happens that a%(d*2^n) can
99 skip a division where (a*2^n)%(d*2^n) can't then there's the same number
100 of divide steps, though how often that happens depends on the assumed
101 distributions of dividend and divisor. In any case this idea is left to
102 CPU specific implementations to consider. */
105 mpn_mod_1_unnorm (mp_srcptr up
, mp_size_t un
, mp_limb_t d
)
117 /* Skip a division if high < divisor. Having the test here before
118 normalizing will still skip as often as possible. */
119 r
= up
[un
- 1] << GMP_NAIL_BITS
;
130 /* If udiv_qrnnd doesn't need a normalized divisor, can use the simple
132 if (! UDIV_NEEDS_NORMALIZATION
133 && BELOW_THRESHOLD (un
, MOD_1_UNNORM_THRESHOLD
))
135 for (i
= un
- 1; i
>= 0; i
--)
137 n0
= up
[i
] << GMP_NAIL_BITS
;
138 udiv_qrnnd (dummy
, r
, r
, n0
, d
);
144 count_leading_zeros (cnt
, d
);
147 n1
= up
[un
- 1] << GMP_NAIL_BITS
;
148 r
= (r
<< cnt
) | (n1
>> (GMP_LIMB_BITS
- cnt
));
150 if (UDIV_NEEDS_NORMALIZATION
151 && BELOW_THRESHOLD (un
, MOD_1_UNNORM_THRESHOLD
))
154 for (i
= un
- 2; i
>= 0; i
--)
156 n0
= up
[i
] << GMP_NAIL_BITS
;
157 nshift
= (n1
<< cnt
) | (n0
>> (GMP_NUMB_BITS
- cnt
));
158 udiv_qrnnd (dummy
, r
, r
, nshift
, d
);
162 udiv_qrnnd (dummy
, r
, r
, n1
<< cnt
, d
);
168 mp_limb_t inv
, nshift
;
169 invert_limb (inv
, d
);
171 for (i
= un
- 2; i
>= 0; i
--)
173 n0
= up
[i
] << GMP_NAIL_BITS
;
174 nshift
= (n1
<< cnt
) | (n0
>> (GMP_NUMB_BITS
- cnt
));
175 udiv_rnnd_preinv (r
, r
, nshift
, d
, inv
);
179 udiv_rnnd_preinv (r
, r
, n1
<< cnt
, d
, inv
);
186 mpn_mod_1_norm (mp_srcptr up
, mp_size_t un
, mp_limb_t d
)
196 ASSERT (d
& GMP_LIMB_HIGHBIT
);
198 /* High limb is initial remainder, possibly with one subtract of
200 r
= up
[un
- 1] << GMP_NAIL_BITS
;
208 if (BELOW_THRESHOLD (un
, MOD_1_NORM_THRESHOLD
))
210 for (i
= un
- 1; i
>= 0; i
--)
212 n0
= up
[i
] << GMP_NAIL_BITS
;
213 udiv_qrnnd (dummy
, r
, r
, n0
, d
);
221 invert_limb (inv
, d
);
222 for (i
= un
- 1; i
>= 0; i
--)
224 n0
= up
[i
] << GMP_NAIL_BITS
;
225 udiv_rnnd_preinv (r
, r
, n0
, d
, inv
);
233 mpn_mod_1 (mp_srcptr ap
, mp_size_t n
, mp_limb_t b
)
238 /* Should this be handled at all? Rely on callers? Note un==0 is currently
239 required by mpz/fdiv_r_ui.c and possibly other places. */
243 if (UNLIKELY ((b
& GMP_NUMB_HIGHBIT
) != 0))
245 if (BELOW_THRESHOLD (n
, MOD_1N_TO_MOD_1_1_THRESHOLD
))
247 return mpn_mod_1_norm (ap
, n
, b
);
252 mpn_mod_1_1p_cps (pre
, b
);
253 return mpn_mod_1_1p (ap
, n
, b
, pre
);
258 if (BELOW_THRESHOLD (n
, MOD_1U_TO_MOD_1_1_THRESHOLD
))
260 return mpn_mod_1_unnorm (ap
, n
, b
);
262 else if (BELOW_THRESHOLD (n
, MOD_1_1_TO_MOD_1_2_THRESHOLD
))
265 mpn_mod_1_1p_cps (pre
, b
);
266 return mpn_mod_1_1p (ap
, n
, b
<< pre
[1], pre
);
268 else if (BELOW_THRESHOLD (n
, MOD_1_2_TO_MOD_1_4_THRESHOLD
) || UNLIKELY (b
> GMP_NUMB_MASK
/ 4))
271 mpn_mod_1s_2p_cps (pre
, b
);
272 return mpn_mod_1s_2p (ap
, n
, b
<< pre
[1], pre
);
277 mpn_mod_1s_4p_cps (pre
, b
);
278 return mpn_mod_1s_4p (ap
, n
, b
<< pre
[1], pre
);