xref: /titanic_50/usr/src/common/bignum/mont_mulf.c (revision 5bc515a198c7f6b3c0036a1b6c10ac9f2119439d)
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 (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  * If compiled without -DRF_INLINE_MACROS then needs -lm at link time
28  * If compiled with -DRF_INLINE_MACROS then needs conv.il at compile time
29  * (i.e. cc <compiler_flags> -DRF_INLINE_MACROS conv.il mont_mulf.c )
30  */
31 
32 #include <sys/types.h>
33 #include <math.h>
34 
35 static const double TwoTo16 = 65536.0;
36 static const double TwoToMinus16 = 1.0/65536.0;
37 static const double Zero = 0.0;
38 static const double TwoTo32 = 65536.0 * 65536.0;
39 static const double TwoToMinus32 = 1.0 / (65536.0 * 65536.0);
40 
41 #ifdef RF_INLINE_MACROS
42 
43 double upper32(double);
44 double lower32(double, double);
45 double mod(double, double, double);
46 
47 #else
48 
49 static double
50 upper32(double x)
51 {
52 	return (floor(x * TwoToMinus32));
53 }
54 
55 
56 static double
57 lower32(double x, double y)
58 {
59 	return (x - TwoTo32 * floor(x * TwoToMinus32));
60 }
61 
62 static double
63 mod(double x, double oneoverm, double m)
64 {
65 	return (x - m * floor(x * oneoverm));
66 }
67 
68 #endif
69 
70 
71 static void
72 cleanup(double *dt, int from, int tlen)
73 {
74 	int i;
75 	double tmp, tmp1, x, x1;
76 
77 	tmp = tmp1 = Zero;
78 
79 	for (i = 2 * from; i < 2 * tlen; i += 2) {
80 		x = dt[i];
81 		x1 = dt[i + 1];
82 		dt[i] = lower32(x, Zero) + tmp;
83 		dt[i + 1] = lower32(x1, Zero) + tmp1;
84 		tmp = upper32(x);
85 		tmp1 = upper32(x1);
86 	}
87 }
88 
89 
90 void
91 conv_d16_to_i32(uint32_t *i32, double *d16, int64_t *tmp, int ilen)
92 {
93 	int i;
94 	int64_t t, t1,		/* Using int64_t and not uint64_t */
95 	    a, b, c, d;		/* because more efficient code is */
96 				/* generated this way, and there  */
97 				/* is no overflow.  */
98 	t1 = 0;
99 	a = (int64_t)d16[0];
100 	b = (int64_t)d16[1];
101 	for (i = 0; i < ilen - 1; i++) {
102 		c = (int64_t)d16[2 * i + 2];
103 		t1 += a & 0xffffffff;
104 		t = (a >> 32);
105 		d = (int64_t)d16[2 * i + 3];
106 		t1 += (b & 0xffff) << 16;
107 		t += (b >> 16) + (t1 >> 32);
108 		i32[i] = t1 & 0xffffffff;
109 		t1 = t;
110 		a = c;
111 		b = d;
112 	}
113 	t1 += a & 0xffffffff;
114 	t = (a >> 32);
115 	t1 += (b & 0xffff) << 16;
116 	i32[i] = t1 & 0xffffffff;
117 }
118 
119 void
120 conv_i32_to_d32(double *d32, uint32_t *i32, int len)
121 {
122 	int i;
123 
124 #pragma pipeloop(0)
125 	for (i = 0; i < len; i++)
126 		d32[i] = (double)(i32[i]);
127 }
128 
129 
130 void
131 conv_i32_to_d16(double *d16, uint32_t *i32, int len)
132 {
133 	int i;
134 	uint32_t a;
135 
136 #pragma pipeloop(0)
137 	for (i = 0; i < len; i++) {
138 		a = i32[i];
139 		d16[2 * i] = (double)(a & 0xffff);
140 		d16[2 * i + 1] = (double)(a >> 16);
141 	}
142 }
143 
144 #ifdef RF_INLINE_MACROS
145 
146 void
147 i16_to_d16_and_d32x4(const double *,	/* 1/(2^16) */
148 			const double *,	/* 2^16 */
149 			const double *,	/* 0 */
150 			double *,	/* result16 */
151 			double *,	/* result32 */
152 			float *);	/* source - should be unsigned int* */
153 					/* converted to float* */
154 
155 #else
156 
157 
158 static void
159 i16_to_d16_and_d32x4(const double *dummy1,	/* 1/(2^16) */
160 			const double *dummy2,	/* 2^16 */
161 			const double *dummy3,	/* 0 */
162 			double *result16,
163 			double *result32,
164 			float *src)	/* source - should be unsigned int* */
165 					/* converted to float* */
166 {
167 	uint32_t *i32;
168 	uint32_t a, b, c, d;
169 
170 	i32 = (uint32_t *)src;
171 	a = i32[0];
172 	b = i32[1];
173 	c = i32[2];
174 	d = i32[3];
175 	result16[0] = (double)(a & 0xffff);
176 	result16[1] = (double)(a >> 16);
177 	result32[0] = (double)a;
178 	result16[2] = (double)(b & 0xffff);
179 	result16[3] = (double)(b >> 16);
180 	result32[1] = (double)b;
181 	result16[4] = (double)(c & 0xffff);
182 	result16[5] = (double)(c >> 16);
183 	result32[2] = (double)c;
184 	result16[6] = (double)(d & 0xffff);
185 	result16[7] = (double)(d >> 16);
186 	result32[3] = (double)d;
187 }
188 
189 #endif
190 
191 
192 void
193 conv_i32_to_d32_and_d16(double *d32, double *d16, uint32_t *i32, int len)
194 {
195 	int i;
196 	uint32_t a;
197 
198 #pragma pipeloop(0)
199 	for (i = 0; i < len - 3; i += 4) {
200 		i16_to_d16_and_d32x4(&TwoToMinus16, &TwoTo16, &Zero,
201 		    &(d16[2*i]), &(d32[i]), (float *)(&(i32[i])));
202 	}
203 	for (; i < len; i++) {
204 		a = i32[i];
205 		d32[i] = (double)(i32[i]);
206 		d16[2 * i] = (double)(a & 0xffff);
207 		d16[2 * i + 1] = (double)(a >> 16);
208 	}
209 }
210 
211 
212 static void
213 adjust_montf_result(uint32_t *i32, uint32_t *nint, int len)
214 {
215 	int64_t acc;
216 	int i;
217 
218 	if (i32[len] > 0)
219 		i = -1;
220 	else {
221 		for (i = len - 1; i >= 0; i--) {
222 			if (i32[i] != nint[i]) break;
223 		}
224 	}
225 	if ((i < 0) || (i32[i] > nint[i])) {
226 		acc = 0;
227 		for (i = 0; i < len; i++) {
228 			acc = acc + (uint64_t)(i32[i]) - (uint64_t)(nint[i]);
229 			i32[i] = acc & 0xffffffff;
230 			acc = acc >> 32;
231 		}
232 	}
233 }
234 
235 
236 /*
237  * the lengths of the input arrays should be at least the following:
238  * result[nlen+1], dm1[nlen], dm2[2*nlen+1], dt[4*nlen+2], dn[nlen], nint[nlen]
239  * all of them should be different from one another
240  */
241 void mont_mulf_noconv(uint32_t *result,
242 			double *dm1, double *dm2, double *dt,
243 			double *dn, uint32_t *nint,
244 			int nlen, double dn0)
245 {
246 	int i, j, jj;
247 	double digit, m2j, a, b;
248 	double *pdm1, *pdm2, *pdn, *pdtj, pdn_0, pdm1_0;
249 
250 	pdm1 = &(dm1[0]);
251 	pdm2 = &(dm2[0]);
252 	pdn = &(dn[0]);
253 	pdm2[2 * nlen] = Zero;
254 
255 	if (nlen != 16) {
256 		for (i = 0; i < 4 * nlen + 2; i++)
257 			dt[i] = Zero;
258 		a = dt[0] = pdm1[0] * pdm2[0];
259 		digit = mod(lower32(a, Zero) * dn0, TwoToMinus16, TwoTo16);
260 
261 		pdtj = &(dt[0]);
262 		for (j = jj = 0; j < 2 * nlen; j++, jj++, pdtj++) {
263 			m2j = pdm2[j];
264 			a = pdtj[0] + pdn[0] * digit;
265 			b = pdtj[1] + pdm1[0] * pdm2[j + 1] + a * TwoToMinus16;
266 			pdtj[1] = b;
267 
268 #pragma pipeloop(0)
269 			for (i = 1; i < nlen; i++) {
270 				pdtj[2 * i] += pdm1[i] * m2j + pdn[i] * digit;
271 			}
272 			if (jj == 30) {
273 				cleanup(dt, j / 2 + 1, 2 * nlen + 1);
274 				jj = 0;
275 			}
276 
277 			digit = mod(lower32(b, Zero) * dn0,
278 			    TwoToMinus16, TwoTo16);
279 		}
280 	} else {
281 		a = dt[0] = pdm1[0] * pdm2[0];
282 
283 		dt[65] = dt[64] = dt[63] = dt[62] = dt[61] = dt[60] =
284 		    dt[59] = dt[58] = dt[57] = dt[56] = dt[55] =
285 		    dt[54] = dt[53] = dt[52] = dt[51] = dt[50] =
286 		    dt[49] = dt[48] = dt[47] = dt[46] = dt[45] =
287 		    dt[44] = dt[43] = dt[42] = dt[41] = dt[40] =
288 		    dt[39] = dt[38] = dt[37] = dt[36] = dt[35] =
289 		    dt[34] = dt[33] = dt[32] = dt[31] = dt[30] =
290 		    dt[29] = dt[28] = dt[27] = dt[26] = dt[25] =
291 		    dt[24] = dt[23] = dt[22] = dt[21] = dt[20] =
292 		    dt[19] = dt[18] = dt[17] = dt[16] = dt[15] =
293 		    dt[14] = dt[13] = dt[12] = dt[11] = dt[10] =
294 		    dt[9] = dt[8] = dt[7] = dt[6] = dt[5] = dt[4] =
295 		    dt[3] = dt[2] = dt[1] = Zero;
296 
297 		pdn_0 = pdn[0];
298 		pdm1_0 = pdm1[0];
299 
300 		digit = mod(lower32(a, Zero) * dn0, TwoToMinus16, TwoTo16);
301 		pdtj = &(dt[0]);
302 
303 		for (j = 0; j < 32; j++, pdtj++) {
304 
305 			m2j = pdm2[j];
306 			a = pdtj[0] + pdn_0 * digit;
307 			b = pdtj[1] + pdm1_0 * pdm2[j + 1] + a * TwoToMinus16;
308 			pdtj[1] = b;
309 
310 			pdtj[2] += pdm1[1] *m2j + pdn[1] * digit;
311 			pdtj[4] += pdm1[2] *m2j + pdn[2] * digit;
312 			pdtj[6] += pdm1[3] *m2j + pdn[3] * digit;
313 			pdtj[8] += pdm1[4] *m2j + pdn[4] * digit;
314 			pdtj[10] += pdm1[5] *m2j + pdn[5] * digit;
315 			pdtj[12] += pdm1[6] *m2j + pdn[6] * digit;
316 			pdtj[14] += pdm1[7] *m2j + pdn[7] * digit;
317 			pdtj[16] += pdm1[8] *m2j + pdn[8] * digit;
318 			pdtj[18] += pdm1[9] *m2j + pdn[9] * digit;
319 			pdtj[20] += pdm1[10] *m2j + pdn[10] * digit;
320 			pdtj[22] += pdm1[11] *m2j + pdn[11] * digit;
321 			pdtj[24] += pdm1[12] *m2j + pdn[12] * digit;
322 			pdtj[26] += pdm1[13] *m2j + pdn[13] * digit;
323 			pdtj[28] += pdm1[14] *m2j + pdn[14] * digit;
324 			pdtj[30] += pdm1[15] *m2j + pdn[15] * digit;
325 			/* no need for cleanup, cannot overflow */
326 			digit = mod(lower32(b, Zero) * dn0,
327 			    TwoToMinus16, TwoTo16);
328 		}
329 	}
330 
331 	conv_d16_to_i32(result, dt + 2 * nlen, (int64_t *)dt, nlen + 1);
332 	adjust_montf_result(result, nint, nlen);
333 }
334