xref: /titanic_50/usr/src/common/bignum/mont_mulf.c (revision 20c794b39650d115e17a15983b6b82e46238cf45)
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  * If compiled without -DRF_INLINE_MACROS then needs -lm at link time
31  * If compiled with -DRF_INLINE_MACROS then needs conv.il at compile time
32  * (i.e. cc <compileer_flags> -DRF_INLINE_MACROS conv.il mont_mulf.c )
33  */
34 
35 #include <sys/types.h>
36 #include <math.h>
37 
38 static const double TwoTo16 = 65536.0;
39 static const double TwoToMinus16 = 1.0/65536.0;
40 static const double Zero = 0.0;
41 static const double TwoTo32 = 65536.0 * 65536.0;
42 static const double TwoToMinus32 = 1.0 / (65536.0 * 65536.0);
43 
44 #ifdef RF_INLINE_MACROS
45 
46 double upper32(double);
47 double lower32(double, double);
48 double mod(double, double, double);
49 
50 #else
51 
52 static double
53 upper32(double x)
54 {
55 	return (floor(x * TwoToMinus32));
56 }
57 
58 
59 static double
60 lower32(double x, double y)
61 {
62 	return (x - TwoTo32 * floor(x * TwoToMinus32));
63 }
64 
65 static double
66 mod(double x, double oneoverm, double m)
67 {
68 	return (x - m * floor(x * oneoverm));
69 }
70 
71 #endif
72 
73 
74 static void
75 cleanup(double *dt, int from, int tlen)
76 {
77 	int i;
78 	double tmp, tmp1, x, x1;
79 
80 	tmp = tmp1 = Zero;
81 
82 	for (i = 2 * from; i < 2 * tlen; i += 2) {
83 		x = dt[i];
84 		x1 = dt[i + 1];
85 		dt[i] = lower32(x, Zero) + tmp;
86 		dt[i + 1] = lower32(x1, Zero) + tmp1;
87 		tmp = upper32(x);
88 		tmp1 = upper32(x1);
89 	}
90 }
91 
92 
93 void
94 conv_d16_to_i32(uint32_t *i32, double *d16, int64_t *tmp, int ilen)
95 {
96 	int i;
97 	int64_t t, t1,		/* using int64_t and not uint64_t */
98 		a, b, c, d;	/* because more efficient code is */
99 				/* generated this way, and there  */
100 				/* is no overflow  */
101 	t1 = 0;
102 	a = (int64_t)d16[0];
103 	b = (int64_t)d16[1];
104 	for (i = 0; i < ilen - 1; i++) {
105 		c = (int64_t)d16[2 * i + 2];
106 		t1 += a & 0xffffffff;
107 		t = (a >> 32);
108 		d = (int64_t)d16[2 * i + 3];
109 		t1 += (b & 0xffff) << 16;
110 		t += (b >> 16) + (t1 >> 32);
111 		i32[i] = t1 & 0xffffffff;
112 		t1 = t;
113 		a = c;
114 		b = d;
115 	}
116 	t1 += a & 0xffffffff;
117 	t = (a >> 32);
118 	t1 += (b & 0xffff) << 16;
119 	i32[i] = t1 & 0xffffffff;
120 }
121 
122 void
123 conv_i32_to_d32(double *d32, uint32_t *i32, int len)
124 {
125 	int i;
126 
127 #pragma pipeloop(0)
128 	for (i = 0; i < len; i++)
129 		d32[i] = (double)(i32[i]);
130 }
131 
132 
133 void
134 conv_i32_to_d16(double *d16, uint32_t *i32, int len)
135 {
136 	int i;
137 	uint32_t a;
138 
139 #pragma pipeloop(0)
140 	for (i = 0; i < len; i++) {
141 		a = i32[i];
142 		d16[2 * i] = (double)(a & 0xffff);
143 		d16[2 * i + 1] = (double)(a >> 16);
144 	}
145 }
146 
147 #ifdef RF_INLINE_MACROS
148 
149 void
150 i16_to_d16_and_d32x4(const double *,	/* 1/(2^16) */
151 			const double *,	/* 2^16 */
152 			const double *,	/* 0 */
153 			double *,	/* result16 */
154 			double *,	/* result32 */
155 			float *);	/* source - should be unsigned int* */
156 					/* converted to float* */
157 
158 #else
159 
160 
161 static void
162 i16_to_d16_and_d32x4(const double *dummy1,	/* 1/(2^16) */
163 			const double *dummy2,	/* 2^16 */
164 			const double *dummy3,	/* 0 */
165 			double *result16,
166 			double *result32,
167 			float *src)	/* source - should be unsigned int* */
168 					/* converted to float* */
169 {
170 	uint32_t *i32;
171 	uint32_t a, b, c, d;
172 
173 	i32 = (uint32_t *)src;
174 	a = i32[0];
175 	b = i32[1];
176 	c = i32[2];
177 	d = i32[3];
178 	result16[0] = (double)(a & 0xffff);
179 	result16[1] = (double)(a >> 16);
180 	result32[0] = (double)a;
181 	result16[2] = (double)(b & 0xffff);
182 	result16[3] = (double)(b >> 16);
183 	result32[1] = (double)b;
184 	result16[4] = (double)(c & 0xffff);
185 	result16[5] = (double)(c >> 16);
186 	result32[2] = (double)c;
187 	result16[6] = (double)(d & 0xffff);
188 	result16[7] = (double)(d >> 16);
189 	result32[3] = (double)d;
190 }
191 
192 #endif
193 
194 
195 void
196 conv_i32_to_d32_and_d16(double *d32, double *d16, uint32_t *i32, int len)
197 {
198 	int i;
199 	uint32_t a;
200 
201 #pragma pipeloop(0)
202 	for (i = 0; i < len - 3; i += 4) {
203 		i16_to_d16_and_d32x4(&TwoToMinus16, &TwoTo16, &Zero,
204 					&(d16[2*i]), &(d32[i]),
205 					(float *)(&(i32[i])));
206 	}
207 	for (; i < len; i++) {
208 		a = i32[i];
209 		d32[i] = (double)(i32[i]);
210 		d16[2 * i] = (double)(a & 0xffff);
211 		d16[2 * i + 1] = (double)(a >> 16);
212 	}
213 }
214 
215 
216 static void
217 adjust_montf_result(uint32_t *i32, uint32_t *nint, int len)
218 {
219 	int64_t acc;
220 	int i;
221 
222 	if (i32[len] > 0)
223 		i = -1;
224 	else {
225 		for (i = len - 1; i >= 0; i--) {
226 			if (i32[i] != nint[i]) break;
227 		}
228 	}
229 	if ((i < 0) || (i32[i] > nint[i])) {
230 		acc = 0;
231 		for (i = 0; i < len; i++) {
232 			acc = acc + (uint64_t)(i32[i]) - (uint64_t)(nint[i]);
233 			i32[i] = acc & 0xffffffff;
234 			acc = acc >> 32;
235 		}
236 	}
237 }
238 
239 
240 /*
241  * the lengths of the input arrays should be at least the following:
242  * result[nlen+1], dm1[nlen], dm2[2*nlen+1], dt[4*nlen+2], dn[nlen], nint[nlen]
243  * all of them should be different from one another
244  */
245 void mont_mulf_noconv(uint32_t *result,
246 			double *dm1, double *dm2, double *dt,
247 			double *dn, uint32_t *nint,
248 			int nlen, double dn0)
249 {
250 	int i, j, jj;
251 	double digit, m2j, a, b;
252 	double *pdm1, *pdm2, *pdn, *pdtj, pdn_0, pdm1_0;
253 
254 	pdm1 = &(dm1[0]);
255 	pdm2 = &(dm2[0]);
256 	pdn = &(dn[0]);
257 	pdm2[2 * nlen] = Zero;
258 
259 	if (nlen != 16) {
260 		for (i = 0; i < 4 * nlen + 2; i++)
261 			dt[i] = Zero;
262 		a = dt[0] = pdm1[0] * pdm2[0];
263 		digit = mod(lower32(a, Zero) * dn0, TwoToMinus16, TwoTo16);
264 
265 		pdtj = &(dt[0]);
266 		for (j = jj = 0; j < 2 * nlen; j++, jj++, pdtj++) {
267 			m2j = pdm2[j];
268 			a = pdtj[0] + pdn[0] * digit;
269 			b = pdtj[1] + pdm1[0] * pdm2[j + 1] + a * TwoToMinus16;
270 			pdtj[1] = b;
271 
272 #pragma pipeloop(0)
273 			for (i = 1; i < nlen; i++) {
274 				pdtj[2 * i] += pdm1[i] * m2j + pdn[i] * digit;
275 			}
276 			if (jj == 30) {
277 				cleanup(dt, j / 2 + 1, 2 * nlen + 1);
278 				jj = 0;
279 			}
280 
281 			digit = mod(lower32(b, Zero) * dn0,
282 				    TwoToMinus16, TwoTo16);
283 		}
284 	} else {
285 		a = dt[0] = pdm1[0] * pdm2[0];
286 
287 		dt[65] = dt[64] = dt[63] = dt[62] = dt[61] = dt[60] =
288 			dt[59] = dt[58] = dt[57] = dt[56] = dt[55] =
289 			dt[54] = dt[53] = dt[52] = dt[51] = dt[50] =
290 			dt[49] = dt[48] = dt[47] = dt[46] = dt[45] =
291 			dt[44] = dt[43] = dt[42] = dt[41] = dt[40] =
292 			dt[39] = dt[38] = dt[37] = dt[36] = dt[35] =
293 			dt[34] = dt[33] = dt[32] = dt[31] = dt[30] =
294 			dt[29] = dt[28] = dt[27] = dt[26] = dt[25] =
295 			dt[24] = dt[23] = dt[22] = dt[21] = dt[20] =
296 			dt[19] = dt[18] = dt[17] = dt[16] = dt[15] =
297 			dt[14] = dt[13] = dt[12] = dt[11] = dt[10] =
298 			dt[9] = dt[8] = dt[7] = dt[6] = dt[5] = dt[4] =
299 			dt[3] = dt[2] = dt[1] = Zero;
300 
301 		pdn_0 = pdn[0];
302 		pdm1_0 = pdm1[0];
303 
304 		digit = mod(lower32(a, Zero) * dn0, TwoToMinus16, TwoTo16);
305 		pdtj = &(dt[0]);
306 
307 		for (j = 0; j < 32; j++, pdtj++) {
308 
309 			m2j = pdm2[j];
310 			a = pdtj[0] + pdn_0 * digit;
311 			b = pdtj[1] + pdm1_0 * pdm2[j + 1] + a * TwoToMinus16;
312 			pdtj[1] = b;
313 
314 			pdtj[2] += pdm1[1] *m2j + pdn[1] * digit;
315 			pdtj[4] += pdm1[2] *m2j + pdn[2] * digit;
316 			pdtj[6] += pdm1[3] *m2j + pdn[3] * digit;
317 			pdtj[8] += pdm1[4] *m2j + pdn[4] * digit;
318 			pdtj[10] += pdm1[5] *m2j + pdn[5] * digit;
319 			pdtj[12] += pdm1[6] *m2j + pdn[6] * digit;
320 			pdtj[14] += pdm1[7] *m2j + pdn[7] * digit;
321 			pdtj[16] += pdm1[8] *m2j + pdn[8] * digit;
322 			pdtj[18] += pdm1[9] *m2j + pdn[9] * digit;
323 			pdtj[20] += pdm1[10] *m2j + pdn[10] * digit;
324 			pdtj[22] += pdm1[11] *m2j + pdn[11] * digit;
325 			pdtj[24] += pdm1[12] *m2j + pdn[12] * digit;
326 			pdtj[26] += pdm1[13] *m2j + pdn[13] * digit;
327 			pdtj[28] += pdm1[14] *m2j + pdn[14] * digit;
328 			pdtj[30] += pdm1[15] *m2j + pdn[15] * digit;
329 			/* no need for cleenup, cannot overflow */
330 			digit = mod(lower32(b, Zero) * dn0,
331 				    TwoToMinus16, TwoTo16);
332 		}
333 	}
334 
335 	conv_d16_to_i32(result, dt + 2 * nlen, (int64_t *)dt, nlen + 1);
336 	adjust_montf_result(result, nint, nlen);
337 }
338