1 /* mpn_toom_interpolate_6pts -- Interpolate for toom43, 52
2
3 Contributed to the GNU project by Marco Bodrato.
4
5 THE FUNCTION IN THIS FILE IS INTERNAL WITH A MUTABLE INTERFACE. IT IS ONLY
6 SAFE TO REACH IT THROUGH DOCUMENTED INTERFACES. IN FACT, IT IS ALMOST
7 GUARANTEED THAT IT WILL CHANGE OR DISAPPEAR IN A FUTURE GNU MP RELEASE.
8
9 Copyright 2009, 2010, 2012 Free Software Foundation, Inc.
10
11 This file is part of the GNU MP Library.
12
13 The GNU MP Library is free software; you can redistribute it and/or modify
14 it under the terms of either:
15
16 * the GNU Lesser General Public License as published by the Free
17 Software Foundation; either version 3 of the License, or (at your
18 option) any later version.
19
20 or
21
22 * the GNU General Public License as published by the Free Software
23 Foundation; either version 2 of the License, or (at your option) any
24 later version.
25
26 or both in parallel, as here.
27
28 The GNU MP Library is distributed in the hope that it will be useful, but
29 WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
30 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
31 for more details.
32
33 You should have received copies of the GNU General Public License and the
34 GNU Lesser General Public License along with the GNU MP Library. If not,
35 see https://www.gnu.org/licenses/. */
36
37 #include "gmp-impl.h"
38
39 #define BINVERT_3 MODLIMB_INVERSE_3
40
41 /* For odd divisors, mpn_divexact_1 works fine with two's complement. */
42 #ifndef mpn_divexact_by3
43 #if HAVE_NATIVE_mpn_pi1_bdiv_q_1
44 #define mpn_divexact_by3(dst,src,size) mpn_pi1_bdiv_q_1(dst,src,size,3,BINVERT_3,0)
45 #else
46 #define mpn_divexact_by3(dst,src,size) mpn_divexact_1(dst,src,size,3)
47 #endif
48 #endif
49
50 /* Interpolation for Toom-3.5, using the evaluation points: infinity,
51 1, -1, 2, -2. More precisely, we want to compute
52 f(2^(GMP_NUMB_BITS * n)) for a polynomial f of degree 5, given the
53 six values
54
55 w5 = f(0),
56 w4 = f(-1),
57 w3 = f(1)
58 w2 = f(-2),
59 w1 = f(2),
60 w0 = limit at infinity of f(x) / x^5,
61
62 The result is stored in {pp, 5*n + w0n}. At entry, w5 is stored at
63 {pp, 2n}, w3 is stored at {pp + 2n, 2n+1}, and w0 is stored at
64 {pp + 5n, w0n}. The other values are 2n + 1 limbs each (with most
65 significant limbs small). f(-1) and f(-2) may be negative, signs
66 determined by the flag bits. All intermediate results are positive.
67 Inputs are destroyed.
68
69 Interpolation sequence was taken from the paper: "Integer and
70 Polynomial Multiplication: Towards Optimal Toom-Cook Matrices".
71 Some slight variations were introduced: adaptation to "gmp
72 instruction set", and a final saving of an operation by interlacing
73 interpolation and recomposition phases.
74 */
75
76 void
mpn_toom_interpolate_6pts(mp_ptr pp,mp_size_t n,enum toom6_flags flags,mp_ptr w4,mp_ptr w2,mp_ptr w1,mp_size_t w0n)77 mpn_toom_interpolate_6pts (mp_ptr pp, mp_size_t n, enum toom6_flags flags,
78 mp_ptr w4, mp_ptr w2, mp_ptr w1,
79 mp_size_t w0n)
80 {
81 mp_limb_t cy;
82 /* cy6 can be stored in w1[2*n], cy4 in w4[0], embankment in w2[0] */
83 mp_limb_t cy4, cy6, embankment;
84
85 ASSERT( n > 0 );
86 ASSERT( 2*n >= w0n && w0n > 0 );
87
88 #define w5 pp /* 2n */
89 #define w3 (pp + 2 * n) /* 2n+1 */
90 #define w0 (pp + 5 * n) /* w0n */
91
92 /* Interpolate with sequence:
93 W2 =(W1 - W2)>>2
94 W1 =(W1 - W5)>>1
95 W1 =(W1 - W2)>>1
96 W4 =(W3 - W4)>>1
97 W2 =(W2 - W4)/3
98 W3 = W3 - W4 - W5
99 W1 =(W1 - W3)/3
100 // Last steps are mixed with recomposition...
101 W2 = W2 - W0<<2
102 W4 = W4 - W2
103 W3 = W3 - W1
104 W2 = W2 - W0
105 */
106
107 /* W2 =(W1 - W2)>>2 */
108 if (flags & toom6_vm2_neg)
109 mpn_add_n (w2, w1, w2, 2 * n + 1);
110 else
111 mpn_sub_n (w2, w1, w2, 2 * n + 1);
112 mpn_rshift (w2, w2, 2 * n + 1, 2);
113
114 /* W1 =(W1 - W5)>>1 */
115 w1[2*n] -= mpn_sub_n (w1, w1, w5, 2*n);
116 mpn_rshift (w1, w1, 2 * n + 1, 1);
117
118 /* W1 =(W1 - W2)>>1 */
119 #if HAVE_NATIVE_mpn_rsh1sub_n
120 mpn_rsh1sub_n (w1, w1, w2, 2 * n + 1);
121 #else
122 mpn_sub_n (w1, w1, w2, 2 * n + 1);
123 mpn_rshift (w1, w1, 2 * n + 1, 1);
124 #endif
125
126 /* W4 =(W3 - W4)>>1 */
127 if (flags & toom6_vm1_neg)
128 {
129 #if HAVE_NATIVE_mpn_rsh1add_n
130 mpn_rsh1add_n (w4, w3, w4, 2 * n + 1);
131 #else
132 mpn_add_n (w4, w3, w4, 2 * n + 1);
133 mpn_rshift (w4, w4, 2 * n + 1, 1);
134 #endif
135 }
136 else
137 {
138 #if HAVE_NATIVE_mpn_rsh1sub_n
139 mpn_rsh1sub_n (w4, w3, w4, 2 * n + 1);
140 #else
141 mpn_sub_n (w4, w3, w4, 2 * n + 1);
142 mpn_rshift (w4, w4, 2 * n + 1, 1);
143 #endif
144 }
145
146 /* W2 =(W2 - W4)/3 */
147 mpn_sub_n (w2, w2, w4, 2 * n + 1);
148 mpn_divexact_by3 (w2, w2, 2 * n + 1);
149
150 /* W3 = W3 - W4 - W5 */
151 mpn_sub_n (w3, w3, w4, 2 * n + 1);
152 w3[2 * n] -= mpn_sub_n (w3, w3, w5, 2 * n);
153
154 /* W1 =(W1 - W3)/3 */
155 mpn_sub_n (w1, w1, w3, 2 * n + 1);
156 mpn_divexact_by3 (w1, w1, 2 * n + 1);
157
158 /*
159 [1 0 0 0 0 0;
160 0 1 0 0 0 0;
161 1 0 1 0 0 0;
162 0 1 0 1 0 0;
163 1 0 1 0 1 0;
164 0 0 0 0 0 1]
165
166 pp[] prior to operations:
167 |_H w0__|_L w0__|______||_H w3__|_L w3__|_H w5__|_L w5__|
168
169 summation scheme for remaining operations:
170 |______________5|n_____4|n_____3|n_____2|n______|n______|pp
171 |_H w0__|_L w0__|______||_H w3__|_L w3__|_H w5__|_L w5__|
172 || H w4 | L w4 |
173 || H w2 | L w2 |
174 || H w1 | L w1 |
175 ||-H w1 |-L w1 |
176 |-H w0 |-L w0 ||-H w2 |-L w2 |
177 */
178 cy = mpn_add_n (pp + n, pp + n, w4, 2 * n + 1);
179 MPN_INCR_U (pp + 3 * n + 1, n, cy);
180
181 /* W2 -= W0<<2 */
182 #if HAVE_NATIVE_mpn_sublsh_n || HAVE_NATIVE_mpn_sublsh2_n_ip1
183 #if HAVE_NATIVE_mpn_sublsh2_n_ip1
184 cy = mpn_sublsh2_n_ip1 (w2, w0, w0n);
185 #else
186 cy = mpn_sublsh_n (w2, w2, w0, w0n, 2);
187 #endif
188 #else
189 /* {W4,2*n+1} is now free and can be overwritten. */
190 cy = mpn_lshift(w4, w0, w0n, 2);
191 cy+= mpn_sub_n(w2, w2, w4, w0n);
192 #endif
193 MPN_DECR_U (w2 + w0n, 2 * n + 1 - w0n, cy);
194
195 /* W4L = W4L - W2L */
196 cy = mpn_sub_n (pp + n, pp + n, w2, n);
197 MPN_DECR_U (w3, 2 * n + 1, cy);
198
199 /* W3H = W3H + W2L */
200 cy4 = w3[2 * n] + mpn_add_n (pp + 3 * n, pp + 3 * n, w2, n);
201 /* W1L + W2H */
202 cy = w2[2 * n] + mpn_add_n (pp + 4 * n, w1, w2 + n, n);
203 MPN_INCR_U (w1 + n, n + 1, cy);
204
205 /* W0 = W0 + W1H */
206 if (LIKELY (w0n > n))
207 cy6 = w1[2 * n] + mpn_add_n (w0, w0, w1 + n, n);
208 else
209 cy6 = mpn_add_n (w0, w0, w1 + n, w0n);
210
211 /*
212 summation scheme for the next operation:
213 |...____5|n_____4|n_____3|n_____2|n______|n______|pp
214 |...w0___|_w1_w2_|_H w3__|_L w3__|_H w5__|_L w5__|
215 ...-w0___|-w1_w2 |
216 */
217 /* if(LIKELY(w0n>n)) the two operands below DO overlap! */
218 cy = mpn_sub_n (pp + 2 * n, pp + 2 * n, pp + 4 * n, n + w0n);
219
220 /* embankment is a "dirty trick" to avoid carry/borrow propagation
221 beyond allocated memory */
222 embankment = w0[w0n - 1] - 1;
223 w0[w0n - 1] = 1;
224 if (LIKELY (w0n > n)) {
225 if (cy4 > cy6)
226 MPN_INCR_U (pp + 4 * n, w0n + n, cy4 - cy6);
227 else
228 MPN_DECR_U (pp + 4 * n, w0n + n, cy6 - cy4);
229 MPN_DECR_U (pp + 3 * n + w0n, 2 * n, cy);
230 MPN_INCR_U (w0 + n, w0n - n, cy6);
231 } else {
232 MPN_INCR_U (pp + 4 * n, w0n + n, cy4);
233 MPN_DECR_U (pp + 3 * n + w0n, 2 * n, cy + cy6);
234 }
235 w0[w0n - 1] += embankment;
236
237 #undef w5
238 #undef w3
239 #undef w0
240
241 }
242