xref: /illumos-gate/usr/src/common/bignum/amd64/bignum_amd64.c (revision 628e3cbed6489fa1db545d8524a06cd6535af456)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*
30  * This file contains bignum implementation code that
31  * is specific to AMD64, but which is still more appropriate
32  * to write in C, rather than assembly language.
33  * bignum_amd64_asm.s does all the assembly language code
34  * for AMD64 specific bignum support.  The assembly language
35  * source file has pure code, no data.  Let the C compiler
36  * generate what is needed to handle the variations in
37  * data representation and addressing, for example,
38  * statically linked vs PIC.
39  */
40 
41 #include "bignum.h"
42 
43 /*
44  * The bignum interface deals only with arrays of 32-bit "digits".
45  * The 64-bit bignum functions are internal implementation details.
46  * If a bignum happens to be aligned on a 64-bit boundary
47  * and its length is even, then the pure 64-bit implementation
48  * can be used.
49  */
50 
51 #define	ISALIGNED64(p) (((uintptr_t)(p) & 7) == 0)
52 #define	ISBIGNUM64(p, len) (ISALIGNED64(p) && (((len) & 1) == 0))
53 
54 #if defined(__lint)
55 
56 extern uint64_t *P64(uint32_t *addr);
57 
58 #else /* lint */
59 
60 #define	P64(addr) ((uint64_t *)addr)
61 
62 #endif /* lint */
63 
64 extern uint64_t big_mul_set_vec64(uint64_t *, uint64_t *, int, uint64_t);
65 extern uint64_t big_mul_add_vec64(uint64_t *, uint64_t *, int, uint64_t);
66 extern void big_mul_vec64(uint64_t *, uint64_t *, int, uint64_t *, int);
67 extern void big_sqr_vec64(uint64_t *, uint64_t *, int);
68 
69 extern uint32_t big_mul_set_vec32(uint32_t *, uint32_t *, int, uint32_t);
70 extern uint32_t big_mul_add_vec32(uint32_t *, uint32_t *, int, uint32_t);
71 extern void big_mul_vec32(uint32_t *, uint32_t *, int, uint32_t *, int);
72 extern void big_sqr_vec32(uint32_t *, uint32_t *, int);
73 
74 uint32_t big_mul_set_vec(uint32_t *, uint32_t *, int, uint32_t);
75 uint32_t big_mul_add_vec(uint32_t *, uint32_t *, int, uint32_t);
76 void big_mul_vec(uint32_t *, uint32_t *, int, uint32_t *, int);
77 void big_sqr_vec(uint32_t *, uint32_t *, int);
78 
79 
80 void
81 big_mul_vec(uint32_t *r, uint32_t *a, int alen, uint32_t *b, int blen)
82 {
83 	if (!ISALIGNED64(r) || !ISBIGNUM64(a, alen) || !ISBIGNUM64(b, blen)) {
84 		big_mul_vec32(r, a, alen, b, blen);
85 		return;
86 	}
87 
88 	big_mul_vec64(P64(r), P64(a), alen / 2, P64(b), blen / 2);
89 }
90 
91 void
92 big_sqr_vec(uint32_t *r, uint32_t *a, int alen)
93 {
94 	if (!ISALIGNED64(r) || !ISBIGNUM64(a, alen)) {
95 		big_mul_vec32(r, a, alen, a, alen);
96 		return;
97 	}
98 	big_sqr_vec64(P64(r), P64(a), alen / 2);
99 }
100 
101 /*
102  * It is OK to cast the 64-bit carry to 32 bit.
103  * There will be no loss, because although we are multiplying the vector, a,
104  * by a uint64_t, its value cannot exceedthat of a uint32_t.
105  */
106 
107 uint32_t
108 big_mul_set_vec(uint32_t *r, uint32_t *a, int alen, uint32_t digit)
109 {
110 	if (!ISALIGNED64(r) || !ISBIGNUM64(a, alen))
111 		return (big_mul_set_vec32(r, a, alen, digit));
112 
113 	return (big_mul_set_vec64(P64(r), P64(a), alen / 2, digit));
114 }
115 uint32_t
116 big_mul_add_vec(uint32_t *r, uint32_t *a, int alen, uint32_t digit)
117 {
118 	if (!ISALIGNED64(r) || !ISBIGNUM64(a, alen))
119 		return (big_mul_add_vec32(r, a, alen, digit));
120 
121 	return (big_mul_add_vec64(P64(r), P64(a), alen / 2, digit));
122 }
123 
124 
125 void
126 big_mul_vec64(uint64_t *r, uint64_t *a, int alen, uint64_t *b, int blen)
127 {
128 	int i;
129 
130 	r[alen] = big_mul_set_vec64(r, a, alen, b[0]);
131 	for (i = 1; i < blen; ++i)
132 		r[alen + i] = big_mul_add_vec64(r+i, a, alen, b[i]);
133 }
134 
135 void
136 big_mul_vec32(uint32_t *r, uint32_t *a, int alen, uint32_t *b, int blen)
137 {
138 	int i;
139 
140 	r[alen] = big_mul_set_vec32(r, a, alen, b[0]);
141 	for (i = 1; i < blen; ++i)
142 		r[alen + i] = big_mul_add_vec32(r+i, a, alen, b[i]);
143 }
144 
145 void
146 big_sqr_vec32(uint32_t *r, uint32_t *a, int alen)
147 {
148 	big_mul_vec32(r, a, alen, a, alen);
149 }
150 
151 
152 uint32_t
153 big_mul_set_vec32(uint32_t *r, uint32_t *a, int alen, uint32_t digit)
154 {
155 	uint64_t p, d, cy;
156 
157 	d = (uint64_t)digit;
158 	cy = 0;
159 	while (alen != 0) {
160 		p = (uint64_t)a[0] * d + cy;
161 		r[0] = (uint32_t)p;
162 		cy = p >> 32;
163 		++r;
164 		++a;
165 		--alen;
166 	}
167 	return ((uint32_t)cy);
168 }
169 
170 uint32_t
171 big_mul_add_vec32(uint32_t *r, uint32_t *a, int alen, uint32_t digit)
172 {
173 	uint64_t p, d, cy;
174 
175 	d = (uint64_t)digit;
176 	cy = 0;
177 	while (alen != 0) {
178 		p = r[0] + (uint64_t)a[0] * d + cy;
179 		r[0] = (uint32_t)p;
180 		cy = p >> 32;
181 		++r;
182 		++a;
183 		--alen;
184 	}
185 	return ((uint32_t)cy);
186 }
187