1 /* UltraSPARC 64 mpn_modexact_1c_odd -- mpn by limb exact style remainder.
2
3 THE FUNCTIONS IN THIS FILE ARE FOR INTERNAL USE ONLY. THEY'RE ALMOST
4 CERTAIN TO BE SUBJECT TO INCOMPATIBLE CHANGES OR DISAPPEAR COMPLETELY IN
5 FUTURE GNU MP RELEASES.
6
7 Copyright 2000-2003 Free Software Foundation, Inc.
8
9 This file is part of the GNU MP Library.
10
11 The GNU MP Library is free software; you can redistribute it and/or modify
12 it under the terms of either:
13
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.
17
18 or
19
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
22 later version.
23
24 or both in parallel, as here.
25
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
29 for more details.
30
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/. */
34
35 #include "gmp-impl.h"
36 #include "longlong.h"
37
38 #include "mpn/sparc64/sparc64.h"
39
40
41 /* 64-bit divisor 32-bit divisor
42 cycles/limb cycles/limb
43 (approx) (approx)
44 Ultrasparc 2i: ? ?
45 */
46
47
48 /* This implementation reduces the number of multiplies done, knowing that
49 on ultrasparc 1 and 2 the mulx instruction stalls the whole chip.
50
51 The key idea is to use the fact that the low limb of q*d equals l, this
52 being the whole purpose of the q calculated. It means there's no need to
53 calculate the lowest 32x32->64 part of the q*d, instead it can be
54 inferred from l and the other three 32x32->64 parts. See sparc64.h for
55 details.
56
57 When d is 32-bits, the same applies, but in this case there's only one
58 other 32x32->64 part (ie. HIGH(q)*d).
59
60 The net effect is that for 64-bit divisor each limb is 4 mulx, or for
61 32-bit divisor each is 2 mulx.
62
63 Enhancements:
64
65 No doubt this could be done in assembler, if that helped the scheduling,
66 or perhaps guaranteed good code irrespective of the compiler.
67
68 Alternatives:
69
70 It might be possibly to use floating point. The loop is dominated by
71 multiply latency, so not sure if floats would improve that. One
72 possibility would be to take two limbs at a time, with a 128 bit inverse,
73 if there's enough registers, which could effectively use float throughput
74 to reduce total latency across two limbs. */
75
76 #define ASSERT_RETVAL(r) \
77 ASSERT (orig_c < d ? r < d : r <= d)
78
79 mp_limb_t
mpn_modexact_1c_odd(mp_srcptr src,mp_size_t size,mp_limb_t d,mp_limb_t orig_c)80 mpn_modexact_1c_odd (mp_srcptr src, mp_size_t size, mp_limb_t d, mp_limb_t orig_c)
81 {
82 mp_limb_t c = orig_c;
83 mp_limb_t s, l, q, h, inverse;
84
85 ASSERT (size >= 1);
86 ASSERT (d & 1);
87 ASSERT_MPN (src, size);
88 ASSERT_LIMB (d);
89 ASSERT_LIMB (c);
90
91 /* udivx is faster than 10 or 12 mulx's for one limb via an inverse */
92 if (size == 1)
93 {
94 s = src[0];
95 if (s > c)
96 {
97 l = s-c;
98 h = l % d;
99 if (h != 0)
100 h = d - h;
101 }
102 else
103 {
104 l = c-s;
105 h = l % d;
106 }
107 return h;
108 }
109
110 binvert_limb (inverse, d);
111
112 if (d <= 0xFFFFFFFF)
113 {
114 s = *src++;
115 size--;
116 do
117 {
118 SUBC_LIMB (c, l, s, c);
119 s = *src++;
120 q = l * inverse;
121 umul_ppmm_half_lowequal (h, q, d, l);
122 c += h;
123 size--;
124 }
125 while (size != 0);
126
127 if (s <= d)
128 {
129 /* With high s <= d the final step can be a subtract and addback.
130 If c==0 then the addback will restore to l>=0. If c==d then
131 will get l==d if s==0, but that's ok per the function
132 definition. */
133
134 l = c - s;
135 l += (l > c ? d : 0);
136
137 ASSERT_RETVAL (l);
138 return l;
139 }
140 else
141 {
142 /* Can't skip a divide, just do the loop code once more. */
143 SUBC_LIMB (c, l, s, c);
144 q = l * inverse;
145 umul_ppmm_half_lowequal (h, q, d, l);
146 c += h;
147
148 ASSERT_RETVAL (c);
149 return c;
150 }
151 }
152 else
153 {
154 mp_limb_t dl = LOW32 (d);
155 mp_limb_t dh = HIGH32 (d);
156 long i;
157
158 s = *src++;
159 size--;
160 do
161 {
162 SUBC_LIMB (c, l, s, c);
163 s = *src++;
164 q = l * inverse;
165 umul_ppmm_lowequal (h, q, d, dh, dl, l);
166 c += h;
167 size--;
168 }
169 while (size != 0);
170
171 if (s <= d)
172 {
173 /* With high s <= d the final step can be a subtract and addback.
174 If c==0 then the addback will restore to l>=0. If c==d then
175 will get l==d if s==0, but that's ok per the function
176 definition. */
177
178 l = c - s;
179 l += (l > c ? d : 0);
180
181 ASSERT_RETVAL (l);
182 return l;
183 }
184 else
185 {
186 /* Can't skip a divide, just do the loop code once more. */
187 SUBC_LIMB (c, l, s, c);
188 q = l * inverse;
189 umul_ppmm_lowequal (h, q, d, dh, dl, l);
190 c += h;
191
192 ASSERT_RETVAL (c);
193 return c;
194 }
195 }
196 }
197