1 /* mulmod_bnm1.c -- multiplication mod B^n-1. 2 3 Contributed to the GNU project by Niels Möller, Torbjorn Granlund and 4 Marco Bodrato. 5 6 THE FUNCTIONS IN THIS FILE ARE INTERNAL WITH MUTABLE INTERFACES. IT IS ONLY 7 SAFE TO REACH THEM THROUGH DOCUMENTED INTERFACES. IN FACT, IT IS ALMOST 8 GUARANTEED THAT THEY WILL CHANGE OR DISAPPEAR IN A FUTURE GNU MP RELEASE. 9 10 Copyright 2009, 2010, 2012, 2013 Free Software Foundation, Inc. 11 12 This file is part of the GNU MP Library. 13 14 The GNU MP Library is free software; you can redistribute it and/or modify 15 it under the terms of either: 16 17 * the GNU Lesser General Public License as published by the Free 18 Software Foundation; either version 3 of the License, or (at your 19 option) any later version. 20 21 or 22 23 * the GNU General Public License as published by the Free Software 24 Foundation; either version 2 of the License, or (at your option) any 25 later version. 26 27 or both in parallel, as here. 28 29 The GNU MP Library is distributed in the hope that it will be useful, but 30 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 31 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 32 for more details. 33 34 You should have received copies of the GNU General Public License and the 35 GNU Lesser General Public License along with the GNU MP Library. If not, 36 see https://www.gnu.org/licenses/. */ 37 38 39 #include "gmp-impl.h" 40 #include "longlong.h" 41 42 /* Inputs are {ap,rn} and {bp,rn}; output is {rp,rn}, computation is 43 mod B^rn - 1, and values are semi-normalised; zero is represented 44 as either 0 or B^n - 1. Needs a scratch of 2rn limbs at tp. 45 tp==rp is allowed. */ 46 void 47 mpn_bc_mulmod_bnm1 (mp_ptr rp, mp_srcptr ap, mp_srcptr bp, mp_size_t rn, 48 mp_ptr tp) 49 { 50 mp_limb_t cy; 51 52 ASSERT (0 < rn); 53 54 mpn_mul_n (tp, ap, bp, rn); 55 cy = mpn_add_n (rp, tp, tp + rn, rn); 56 /* If cy == 1, then the value of rp is at most B^rn - 2, so there can 57 * be no overflow when adding in the carry. */ 58 MPN_INCR_U (rp, rn, cy); 59 } 60 61 62 /* Inputs are {ap,rn+1} and {bp,rn+1}; output is {rp,rn+1}, in 63 semi-normalised representation, computation is mod B^rn + 1. Needs 64 a scratch area of 2rn + 2 limbs at tp; tp == rp is allowed. 65 Output is normalised. */ 66 static void 67 mpn_bc_mulmod_bnp1 (mp_ptr rp, mp_srcptr ap, mp_srcptr bp, mp_size_t rn, 68 mp_ptr tp) 69 { 70 mp_limb_t cy; 71 72 ASSERT (0 < rn); 73 74 mpn_mul_n (tp, ap, bp, rn + 1); 75 ASSERT (tp[2*rn+1] == 0); 76 ASSERT (tp[2*rn] < GMP_NUMB_MAX); 77 cy = tp[2*rn] + mpn_sub_n (rp, tp, tp+rn, rn); 78 rp[rn] = 0; 79 MPN_INCR_U (rp, rn+1, cy); 80 } 81 82 83 /* Computes {rp,MIN(rn,an+bn)} <- {ap,an}*{bp,bn} Mod(B^rn-1) 84 * 85 * The result is expected to be ZERO if and only if one of the operand 86 * already is. Otherwise the class [0] Mod(B^rn-1) is represented by 87 * B^rn-1. This should not be a problem if mulmod_bnm1 is used to 88 * combine results and obtain a natural number when one knows in 89 * advance that the final value is less than (B^rn-1). 90 * Moreover it should not be a problem if mulmod_bnm1 is used to 91 * compute the full product with an+bn <= rn, because this condition 92 * implies (B^an-1)(B^bn-1) < (B^rn-1) . 93 * 94 * Requires 0 < bn <= an <= rn and an + bn > rn/2 95 * Scratch need: rn + (need for recursive call OR rn + 4). This gives 96 * 97 * S(n) <= rn + MAX (rn + 4, S(n/2)) <= 2rn + 4 98 */ 99 void 100 mpn_mulmod_bnm1 (mp_ptr rp, mp_size_t rn, mp_srcptr ap, mp_size_t an, mp_srcptr bp, mp_size_t bn, mp_ptr tp) 101 { 102 ASSERT (0 < bn); 103 ASSERT (bn <= an); 104 ASSERT (an <= rn); 105 106 if ((rn & 1) != 0 || BELOW_THRESHOLD (rn, MULMOD_BNM1_THRESHOLD)) 107 { 108 if (UNLIKELY (bn < rn)) 109 { 110 if (UNLIKELY (an + bn <= rn)) 111 { 112 mpn_mul (rp, ap, an, bp, bn); 113 } 114 else 115 { 116 mp_limb_t cy; 117 mpn_mul (tp, ap, an, bp, bn); 118 cy = mpn_add (rp, tp, rn, tp + rn, an + bn - rn); 119 MPN_INCR_U (rp, rn, cy); 120 } 121 } 122 else 123 mpn_bc_mulmod_bnm1 (rp, ap, bp, rn, tp); 124 } 125 else 126 { 127 mp_size_t n; 128 mp_limb_t cy; 129 mp_limb_t hi; 130 131 n = rn >> 1; 132 133 /* We need at least an + bn >= n, to be able to fit one of the 134 recursive products at rp. Requiring strict inequality makes 135 the code slightly simpler. If desired, we could avoid this 136 restriction by initially halving rn as long as rn is even and 137 an + bn <= rn/2. */ 138 139 ASSERT (an + bn > n); 140 141 /* Compute xm = a*b mod (B^n - 1), xp = a*b mod (B^n + 1) 142 and crt together as 143 144 x = -xp * B^n + (B^n + 1) * [ (xp + xm)/2 mod (B^n-1)] 145 */ 146 147 #define a0 ap 148 #define a1 (ap + n) 149 #define b0 bp 150 #define b1 (bp + n) 151 152 #define xp tp /* 2n + 2 */ 153 /* am1 maybe in {xp, n} */ 154 /* bm1 maybe in {xp + n, n} */ 155 #define sp1 (tp + 2*n + 2) 156 /* ap1 maybe in {sp1, n + 1} */ 157 /* bp1 maybe in {sp1 + n + 1, n + 1} */ 158 159 { 160 mp_srcptr am1, bm1; 161 mp_size_t anm, bnm; 162 mp_ptr so; 163 164 bm1 = b0; 165 bnm = bn; 166 if (LIKELY (an > n)) 167 { 168 am1 = xp; 169 cy = mpn_add (xp, a0, n, a1, an - n); 170 MPN_INCR_U (xp, n, cy); 171 anm = n; 172 so = xp + n; 173 if (LIKELY (bn > n)) 174 { 175 bm1 = so; 176 cy = mpn_add (so, b0, n, b1, bn - n); 177 MPN_INCR_U (so, n, cy); 178 bnm = n; 179 so += n; 180 } 181 } 182 else 183 { 184 so = xp; 185 am1 = a0; 186 anm = an; 187 } 188 189 mpn_mulmod_bnm1 (rp, n, am1, anm, bm1, bnm, so); 190 } 191 192 { 193 int k; 194 mp_srcptr ap1, bp1; 195 mp_size_t anp, bnp; 196 197 bp1 = b0; 198 bnp = bn; 199 if (LIKELY (an > n)) { 200 ap1 = sp1; 201 cy = mpn_sub (sp1, a0, n, a1, an - n); 202 sp1[n] = 0; 203 MPN_INCR_U (sp1, n + 1, cy); 204 anp = n + ap1[n]; 205 if (LIKELY (bn > n)) { 206 bp1 = sp1 + n + 1; 207 cy = mpn_sub (sp1 + n + 1, b0, n, b1, bn - n); 208 sp1[2*n+1] = 0; 209 MPN_INCR_U (sp1 + n + 1, n + 1, cy); 210 bnp = n + bp1[n]; 211 } 212 } else { 213 ap1 = a0; 214 anp = an; 215 } 216 217 if (BELOW_THRESHOLD (n, MUL_FFT_MODF_THRESHOLD)) 218 k=0; 219 else 220 { 221 int mask; 222 k = mpn_fft_best_k (n, 0); 223 mask = (1<<k) - 1; 224 while (n & mask) {k--; mask >>=1;}; 225 } 226 if (k >= FFT_FIRST_K) 227 xp[n] = mpn_mul_fft (xp, n, ap1, anp, bp1, bnp, k); 228 else if (UNLIKELY (bp1 == b0)) 229 { 230 ASSERT (anp + bnp <= 2*n+1); 231 ASSERT (anp + bnp > n); 232 ASSERT (anp >= bnp); 233 mpn_mul (xp, ap1, anp, bp1, bnp); 234 anp = anp + bnp - n; 235 ASSERT (anp <= n || xp[2*n]==0); 236 anp-= anp > n; 237 cy = mpn_sub (xp, xp, n, xp + n, anp); 238 xp[n] = 0; 239 MPN_INCR_U (xp, n+1, cy); 240 } 241 else 242 mpn_bc_mulmod_bnp1 (xp, ap1, bp1, n, xp); 243 } 244 245 /* Here the CRT recomposition begins. 246 247 xm <- (xp + xm)/2 = (xp + xm)B^n/2 mod (B^n-1) 248 Division by 2 is a bitwise rotation. 249 250 Assumes xp normalised mod (B^n+1). 251 252 The residue class [0] is represented by [B^n-1]; except when 253 both input are ZERO. 254 */ 255 256 #if HAVE_NATIVE_mpn_rsh1add_n || HAVE_NATIVE_mpn_rsh1add_nc 257 #if HAVE_NATIVE_mpn_rsh1add_nc 258 cy = mpn_rsh1add_nc(rp, rp, xp, n, xp[n]); /* B^n = 1 */ 259 hi = cy << (GMP_NUMB_BITS - 1); 260 cy = 0; 261 /* next update of rp[n-1] will set cy = 1 only if rp[n-1]+=hi 262 overflows, i.e. a further increment will not overflow again. */ 263 #else /* ! _nc */ 264 cy = xp[n] + mpn_rsh1add_n(rp, rp, xp, n); /* B^n = 1 */ 265 hi = (cy<<(GMP_NUMB_BITS-1))&GMP_NUMB_MASK; /* (cy&1) << ... */ 266 cy >>= 1; 267 /* cy = 1 only if xp[n] = 1 i.e. {xp,n} = ZERO, this implies that 268 the rsh1add was a simple rshift: the top bit is 0. cy=1 => hi=0. */ 269 #endif 270 #if GMP_NAIL_BITS == 0 271 add_ssaaaa(cy, rp[n-1], cy, rp[n-1], 0, hi); 272 #else 273 cy += (hi & rp[n-1]) >> (GMP_NUMB_BITS-1); 274 rp[n-1] ^= hi; 275 #endif 276 #else /* ! HAVE_NATIVE_mpn_rsh1add_n */ 277 #if HAVE_NATIVE_mpn_add_nc 278 cy = mpn_add_nc(rp, rp, xp, n, xp[n]); 279 #else /* ! _nc */ 280 cy = xp[n] + mpn_add_n(rp, rp, xp, n); /* xp[n] == 1 implies {xp,n} == ZERO */ 281 #endif 282 cy += (rp[0]&1); 283 mpn_rshift(rp, rp, n, 1); 284 ASSERT (cy <= 2); 285 hi = (cy<<(GMP_NUMB_BITS-1))&GMP_NUMB_MASK; /* (cy&1) << ... */ 286 cy >>= 1; 287 /* We can have cy != 0 only if hi = 0... */ 288 ASSERT ((rp[n-1] & GMP_NUMB_HIGHBIT) == 0); 289 rp[n-1] |= hi; 290 /* ... rp[n-1] + cy can not overflow, the following INCR is correct. */ 291 #endif 292 ASSERT (cy <= 1); 293 /* Next increment can not overflow, read the previous comments about cy. */ 294 ASSERT ((cy == 0) || ((rp[n-1] & GMP_NUMB_HIGHBIT) == 0)); 295 MPN_INCR_U(rp, n, cy); 296 297 /* Compute the highest half: 298 ([(xp + xm)/2 mod (B^n-1)] - xp ) * B^n 299 */ 300 if (UNLIKELY (an + bn < rn)) 301 { 302 /* Note that in this case, the only way the result can equal 303 zero mod B^{rn} - 1 is if one of the inputs is zero, and 304 then the output of both the recursive calls and this CRT 305 reconstruction is zero, not B^{rn} - 1. Which is good, 306 since the latter representation doesn't fit in the output 307 area.*/ 308 cy = mpn_sub_n (rp + n, rp, xp, an + bn - n); 309 310 /* FIXME: This subtraction of the high parts is not really 311 necessary, we do it to get the carry out, and for sanity 312 checking. */ 313 cy = xp[n] + mpn_sub_nc (xp + an + bn - n, rp + an + bn - n, 314 xp + an + bn - n, rn - (an + bn), cy); 315 ASSERT (an + bn == rn - 1 || 316 mpn_zero_p (xp + an + bn - n + 1, rn - 1 - (an + bn))); 317 cy = mpn_sub_1 (rp, rp, an + bn, cy); 318 ASSERT (cy == (xp + an + bn - n)[0]); 319 } 320 else 321 { 322 cy = xp[n] + mpn_sub_n (rp + n, rp, xp, n); 323 /* cy = 1 only if {xp,n+1} is not ZERO, i.e. {rp,n} is not ZERO. 324 DECR will affect _at most_ the lowest n limbs. */ 325 MPN_DECR_U (rp, 2*n, cy); 326 } 327 #undef a0 328 #undef a1 329 #undef b0 330 #undef b1 331 #undef xp 332 #undef sp1 333 } 334 } 335 336 mp_size_t 337 mpn_mulmod_bnm1_next_size (mp_size_t n) 338 { 339 mp_size_t nh; 340 341 if (BELOW_THRESHOLD (n, MULMOD_BNM1_THRESHOLD)) 342 return n; 343 if (BELOW_THRESHOLD (n, 4 * (MULMOD_BNM1_THRESHOLD - 1) + 1)) 344 return (n + (2-1)) & (-2); 345 if (BELOW_THRESHOLD (n, 8 * (MULMOD_BNM1_THRESHOLD - 1) + 1)) 346 return (n + (4-1)) & (-4); 347 348 nh = (n + 1) >> 1; 349 350 if (BELOW_THRESHOLD (nh, MUL_FFT_MODF_THRESHOLD)) 351 return (n + (8-1)) & (-8); 352 353 return 2 * mpn_fft_next_size (nh, mpn_fft_best_k (nh, 0)); 354 } 355