xref: /illumos-gate/usr/src/lib/libc/sparc/fp/__quad_mag.c (revision 7a6d80f1660abd4755c68cbd094d4a914681d26e)
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 (c) 1994-1997, by Sun Microsystems, Inc.
24  * All rights reserved.
25  */
26 
27 /*
28  * This file contains __quad_mag_add and __quad_mag_sub, the core
29  * of the quad precision add and subtract operations.
30  */
31 
32 #include "quad.h"
33 
34 /*
35  * __quad_mag_add(x, y, z, fsr)
36  *
37  * Sets *z = *x + *y, rounded according to the rounding mode in *fsr,
38  * and updates the current exceptions in *fsr.  This routine assumes
39  * *x and *y are finite, with the same sign (i.e., an addition of
40  * magnitudes), |*x| >= |*y|, and *z already has its sign bit set.
41  */
42 void
43 __quad_mag_add(const union longdouble *x, const union longdouble *y,
44 	union longdouble *z, unsigned int *fsr)
45 {
46 	unsigned int	lx, ly, ex, ey, frac2, frac3, frac4;
47 	unsigned int	round, sticky, carry, rm;
48 	int		e, uflo;
49 
50 	/* get the leading significand words and exponents */
51 	ex = (x->l.msw & 0x7fffffff) >> 16;
52 	lx = x->l.msw & 0xffff;
53 	if (ex == 0)
54 		ex = 1;
55 	else
56 		lx |= 0x10000;
57 
58 	ey = (y->l.msw & 0x7fffffff) >> 16;
59 	ly = y->l.msw & 0xffff;
60 	if (ey == 0)
61 		ey = 1;
62 	else
63 		ly |= 0x10000;
64 
65 	/* prenormalize y */
66 	e = (int) ex - (int) ey;
67 	round = sticky = 0;
68 	if (e >= 114) {
69 		frac2 = x->l.frac2;
70 		frac3 = x->l.frac3;
71 		frac4 = x->l.frac4;
72 		sticky = ly | y->l.frac2 | y->l.frac3 | y->l.frac4;
73 	} else {
74 		frac2 = y->l.frac2;
75 		frac3 = y->l.frac3;
76 		frac4 = y->l.frac4;
77 		if (e >= 96) {
78 			sticky = frac4 | frac3 | (frac2 & 0x7fffffff);
79 			round = frac2 & 0x80000000;
80 			frac4 = ly;
81 			frac3 = frac2 = ly = 0;
82 			e -= 96;
83 		} else if (e >= 64) {
84 			sticky = frac4 | (frac3 & 0x7fffffff);
85 			round = frac3 & 0x80000000;
86 			frac4 = frac2;
87 			frac3 = ly;
88 			frac2 = ly = 0;
89 			e -= 64;
90 		} else if (e >= 32) {
91 			sticky = frac4 & 0x7fffffff;
92 			round = frac4 & 0x80000000;
93 			frac4 = frac3;
94 			frac3 = frac2;
95 			frac2 = ly;
96 			ly = 0;
97 			e -= 32;
98 		}
99 		if (e) {
100 			sticky |= round | (frac4 & ((1 << (e - 1)) - 1));
101 			round = frac4 & (1 << (e - 1));
102 			frac4 = (frac4 >> e) | (frac3 << (32 - e));
103 			frac3 = (frac3 >> e) | (frac2 << (32 - e));
104 			frac2 = (frac2 >> e) | (ly << (32 - e));
105 			ly >>= e;
106 		}
107 
108 		/* add, propagating carries */
109 		frac4 += x->l.frac4;
110 		carry = (frac4 < x->l.frac4);
111 		frac3 += x->l.frac3;
112 		if (carry) {
113 			frac3++;
114 			carry = (frac3 <= x->l.frac3);
115 		} else {
116 			carry = (frac3 < x->l.frac3);
117 		}
118 		frac2 += x->l.frac2;
119 		if (carry) {
120 			frac2++;
121 			carry = (frac2 <= x->l.frac2);
122 		} else {
123 			carry = (frac2 < x->l.frac2);
124 		}
125 		lx += ly;
126 		if (carry)
127 			lx++;
128 
129 		/* postnormalize */
130 		if (lx >= 0x20000) {
131 			sticky |= round;
132 			round = frac4 & 1;
133 			frac4 = (frac4 >> 1) | (frac3 << 31);
134 			frac3 = (frac3 >> 1) | (frac2 << 31);
135 			frac2 = (frac2 >> 1) | (lx << 31);
136 			lx >>= 1;
137 			ex++;
138 		}
139 	}
140 
141 	/* keep track of whether the result before rounding is tiny */
142 	uflo = (lx < 0x10000);
143 
144 	/* get the rounding mode, fudging directed rounding modes */
145 	/* as though the result were positive */
146 	rm = *fsr >> 30;
147 	if (z->l.msw)
148 		rm ^= (rm >> 1);
149 
150 	/* see if we need to round */
151 	if (round | sticky) {
152 		*fsr |= FSR_NXC;
153 
154 		/* round up if necessary */
155 		if (rm == FSR_RP || (rm == FSR_RN && round &&
156 			(sticky || (frac4 & 1)))) {
157 			if (++frac4 == 0)
158 				if (++frac3 == 0)
159 					if (++frac2 == 0)
160 						if (++lx >= 0x20000) {
161 							lx >>= 1;
162 							ex++;
163 						}
164 		}
165 	}
166 
167 	/* check for overflow */
168 	if (ex >= 0x7fff) {
169 		/* store the default overflowed result */
170 		*fsr |= FSR_OFC | FSR_NXC;
171 		if (rm == FSR_RN || rm == FSR_RP) {
172 			z->l.msw |= 0x7fff0000;
173 			z->l.frac2 = z->l.frac3 = z->l.frac4 = 0;
174 		} else {
175 			z->l.msw |= 0x7ffeffff;
176 			z->l.frac2 = z->l.frac3 = z->l.frac4 = 0xffffffff;
177 		}
178 	} else {
179 		/* store the result */
180 		if (lx >= 0x10000)
181 			z->l.msw |= (ex << 16);
182 		z->l.msw |= (lx & 0xffff);
183 		z->l.frac2 = frac2;
184 		z->l.frac3 = frac3;
185 		z->l.frac4 = frac4;
186 
187 		/* if the pre-rounded result was tiny and underflow trapping */
188 		/* is enabled, simulate underflow */
189 		if (uflo && (*fsr & FSR_UFM))
190 			*fsr |= FSR_UFC;
191 	}
192 }
193 
194 /*
195  * __quad_mag_sub(x, y, z, fsr)
196  *
197  * Sets *z = *x - *y, rounded according to the rounding mode in *fsr,
198  * and updates the current exceptions in *fsr.  This routine assumes
199  * *x and *y are finite, with opposite signs (i.e., a subtraction of
200  * magnitudes), |*x| >= |*y|, and *z already has its sign bit set.
201  */
202 void
203 __quad_mag_sub(const union longdouble *x, const union longdouble *y,
204 	union longdouble *z, unsigned int *fsr)
205 {
206 	unsigned int	lx, ly, ex, ey, frac2, frac3, frac4;
207 	unsigned int	guard, round, sticky, borrow, rm;
208 	int		e;
209 
210 	/* get the leading significand words and exponents */
211 	ex = (x->l.msw & 0x7fffffff) >> 16;
212 	lx = x->l.msw & 0xffff;
213 	if (ex == 0)
214 		ex = 1;
215 	else
216 		lx |= 0x10000;
217 
218 	ey = (y->l.msw & 0x7fffffff) >> 16;
219 	ly = y->l.msw & 0xffff;
220 	if (ey == 0)
221 		ey = 1;
222 	else
223 		ly |= 0x10000;
224 
225 	/* prenormalize y */
226 	e = (int) ex - (int) ey;
227 	guard = round = sticky = 0;
228 	if (e > 114) {
229 		sticky = ly | y->l.frac2 | y->l.frac3 | y->l.frac4;
230 		ly = frac2 = frac3 = frac4 = 0;
231 	} else {
232 		frac2 = y->l.frac2;
233 		frac3 = y->l.frac3;
234 		frac4 = y->l.frac4;
235 		if (e >= 96) {
236 			sticky = frac4 | frac3 | (frac2 & 0x3fffffff);
237 			round = frac2 & 0x40000000;
238 			guard = frac2 & 0x80000000;
239 			frac4 = ly;
240 			frac3 = frac2 = ly = 0;
241 			e -= 96;
242 		} else if (e >= 64) {
243 			sticky = frac4 | (frac3 & 0x3fffffff);
244 			round = frac3 & 0x40000000;
245 			guard = frac3 & 0x80000000;
246 			frac4 = frac2;
247 			frac3 = ly;
248 			frac2 = ly = 0;
249 			e -= 64;
250 		} else if (e >= 32) {
251 			sticky = frac4 & 0x3fffffff;
252 			round = frac4 & 0x40000000;
253 			guard = frac4 & 0x80000000;
254 			frac4 = frac3;
255 			frac3 = frac2;
256 			frac2 = ly;
257 			ly = 0;
258 			e -= 32;
259 		}
260 		if (e > 1) {
261 			sticky |= guard | round |
262 				(frac4 & ((1 << (e - 2)) - 1));
263 			round = frac4 & (1 << (e - 2));
264 			guard = frac4 & (1 << (e - 1));
265 			frac4 = (frac4 >> e) | (frac3 << (32 - e));
266 			frac3 = (frac3 >> e) | (frac2 << (32 - e));
267 			frac2 = (frac2 >> e) | (ly << (32 - e));
268 			ly >>= e;
269 		} else if (e == 1) {
270 			sticky |= round;
271 			round = guard;
272 			guard = frac4 & 1;
273 			frac4 = (frac4 >> 1) | (frac3 << 31);
274 			frac3 = (frac3 >> 1) | (frac2 << 31);
275 			frac2 = (frac2 >> 1) | (ly << 31);
276 			ly >>= 1;
277 		}
278 	}
279 
280 	/* complement guard, round, and sticky as need be */
281 	if (sticky) {
282 		round = !round;
283 		guard = !guard;
284 	} else if (round) {
285 		guard = !guard;
286 	}
287 	borrow = (guard | round | sticky);
288 
289 	/* subtract, propagating borrows */
290 	frac4 = x->l.frac4 - frac4;
291 	if (borrow) {
292 		frac4--;
293 		borrow = (frac4 >= x->l.frac4);
294 	} else {
295 		borrow = (frac4 > x->l.frac4);
296 	}
297 	frac3 = x->l.frac3 - frac3;
298 	if (borrow) {
299 		frac3--;
300 		borrow = (frac3 >= x->l.frac3);
301 	} else {
302 		borrow = (frac3 > x->l.frac3);
303 	}
304 	frac2 = x->l.frac2 - frac2;
305 	if (borrow) {
306 		frac2--;
307 		borrow = (frac2 >= x->l.frac2);
308 	} else {
309 		borrow = (frac2 > x->l.frac2);
310 	}
311 	lx -= ly;
312 	if (borrow)
313 		lx--;
314 
315 	/* get the rounding mode */
316 	rm = *fsr >> 30;
317 
318 	/* handle zero result */
319 	if (!(lx | frac2 | frac3 | frac4 | guard)) {
320 		z->l.msw = ((rm == FSR_RM)? 0x80000000 : 0);
321 		z->l.frac2 = z->l.frac3 = z->l.frac4 = 0;
322 		return;
323 	}
324 
325 	/* postnormalize */
326 	if (lx < 0x10000) {
327 		/* if cancellation occurred or the exponent is 1, */
328 		/* the result is exact */
329 		if (lx < 0x8000 || ex == 1) {
330 			while ((lx | (frac2 & 0xfffe0000)) == 0 && ex > 32) {
331 				lx = frac2;
332 				frac2 = frac3;
333 				frac3 = frac4;
334 				frac4 = ((guard)? 0x80000000 : 0);
335 				guard = 0;
336 				ex -= 32;
337 			}
338 			while (lx < 0x10000 && ex > 1) {
339 				lx = (lx << 1) | (frac2 >> 31);
340 				frac2 = (frac2 << 1) | (frac3 >> 31);
341 				frac3 = (frac3 << 1) | (frac4 >> 31);
342 				frac4 <<= 1;
343 				if (guard) {
344 					frac4 |= 1;
345 					guard = 0;
346 				}
347 				ex--;
348 			}
349 			if (lx >= 0x10000)
350 				z->l.msw |= (ex << 16);
351 			z->l.msw |= (lx & 0xffff);
352 			z->l.frac2 = frac2;
353 			z->l.frac3 = frac3;
354 			z->l.frac4 = frac4;
355 
356 			/* if the result is tiny and underflow trapping is */
357 			/* enabled, simulate underflow */
358 			if (lx < 0x10000 && (*fsr & FSR_UFM))
359 				*fsr |= FSR_UFC;
360 			return;
361 		}
362 
363 		/* otherwise we only borrowed one place */
364 		lx = (lx << 1) | (frac2 >> 31);
365 		frac2 = (frac2 << 1) | (frac3 >> 31);
366 		frac3 = (frac3 << 1) | (frac4 >> 31);
367 		frac4 <<= 1;
368 		if (guard)
369 			frac4 |= 1;
370 		ex--;
371 	} else {
372 		sticky |= round;
373 		round = guard;
374 	}
375 
376 	/* fudge directed rounding modes as though the result were positive */
377 	if (z->l.msw)
378 		rm ^= (rm >> 1);
379 
380 	/* see if we need to round */
381 	if (round | sticky) {
382 		*fsr |= FSR_NXC;
383 
384 		/* round up if necessary */
385 		if (rm == FSR_RP || (rm == FSR_RN && round &&
386 			(sticky || (frac4 & 1)))) {
387 			if (++frac4 == 0)
388 				if (++frac3 == 0)
389 					if (++frac2 == 0)
390 						if (++lx >= 0x20000) {
391 							lx >>= 1;
392 							ex++;
393 						}
394 		}
395 	}
396 
397 	/* store the result */
398 	z->l.msw |= (ex << 16) | (lx & 0xffff);
399 	z->l.frac2 = frac2;
400 	z->l.frac3 = frac3;
401 	z->l.frac4 = frac4;
402 }
403