xref: /netbsd-src/sys/arch/hppa/spmath/dfadd.c (revision b179d0ed445a7ffce6cb8fcf02d64f401be5616c)
1 /*	$NetBSD: dfadd.c,v 1.7 2023/07/31 20:48:04 andvar Exp $	*/
2 
3 /*	$OpenBSD: dfadd.c,v 1.4 2001/03/29 03:58:17 mickey Exp $	*/
4 
5 /*
6  * Copyright 1996 1995 by Open Software Foundation, Inc.
7  *              All Rights Reserved
8  *
9  * Permission to use, copy, modify, and distribute this software and
10  * its documentation for any purpose and without fee is hereby granted,
11  * provided that the above copyright notice appears in all copies and
12  * that both the copyright notice and this permission notice appear in
13  * supporting documentation.
14  *
15  * OSF DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE
16  * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
17  * FOR A PARTICULAR PURPOSE.
18  *
19  * IN NO EVENT SHALL OSF BE LIABLE FOR ANY SPECIAL, INDIRECT, OR
20  * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
21  * LOSS OF USE, DATA OR PROFITS, WHETHER IN ACTION OF CONTRACT,
22  * NEGLIGENCE, OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION
23  * WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
24  */
25 /*
26  * pmk1.1
27  */
28 /*
29  * (c) Copyright 1986 HEWLETT-PACKARD COMPANY
30  *
31  * To anyone who acknowledges that this file is provided "AS IS"
32  * without any express or implied warranty:
33  *     permission to use, copy, modify, and distribute this file
34  * for any purpose is hereby granted without fee, provided that
35  * the above copyright notice and this notice appears in all
36  * copies, and that the name of Hewlett-Packard Company not be
37  * used in advertising or publicity pertaining to distribution
38  * of the software without specific, written prior permission.
39  * Hewlett-Packard Company makes no representations about the
40  * suitability of this software for any purpose.
41  */
42 
43 #include <sys/cdefs.h>
44 __KERNEL_RCSID(0, "$NetBSD: dfadd.c,v 1.7 2023/07/31 20:48:04 andvar Exp $");
45 
46 #include "../spmath/float.h"
47 #include "../spmath/dbl_float.h"
48 
49 /*
50  * Double_add: add two double precision values.
51  */
52 int
dbl_fadd(dbl_floating_point * leftptr,dbl_floating_point * rightptr,dbl_floating_point * dstptr,unsigned int * status)53 dbl_fadd(dbl_floating_point *leftptr, dbl_floating_point *rightptr,
54     dbl_floating_point *dstptr, unsigned int *status)
55 {
56     register unsigned int signless_upper_left, signless_upper_right, save;
57     register unsigned int leftp1, leftp2, rightp1, rightp2, extent;
58     register unsigned int resultp1 = 0, resultp2 = 0;
59 
60     register int result_exponent, right_exponent, diff_exponent;
61     register int sign_save, jumpsize;
62     register int inexact = false;
63     register int underflowtrap;
64 
65     /* Create local copies of the numbers */
66     Dbl_copyfromptr(leftptr,leftp1,leftp2);
67     Dbl_copyfromptr(rightptr,rightp1,rightp2);
68 
69     /* A zero "save" helps discover equal operands (for later),	*
70      * and is used in swapping operands (if needed).		*/
71     Dbl_xortointp1(leftp1,rightp1,/*to*/save);
72 
73     /*
74      * check first operand for NaN's or infinity
75      */
76     if ((result_exponent = Dbl_exponent(leftp1)) == DBL_INFINITY_EXPONENT)
77 	{
78 	if (Dbl_iszero_mantissa(leftp1,leftp2))
79 	    {
80 	    if (Dbl_isnotnan(rightp1,rightp2))
81 		{
82 		if (Dbl_isinfinity(rightp1,rightp2) && save!=0)
83 		    {
84 		    /*
85 		     * invalid since operands are opposite signed infinity's
86 		     */
87 		    if (Is_invalidtrap_enabled()) return(INVALIDEXCEPTION);
88 		    Set_invalidflag();
89 		    Dbl_makequietnan(resultp1,resultp2);
90 		    Dbl_copytoptr(resultp1,resultp2,dstptr);
91 		    return(NOEXCEPTION);
92 		    }
93 		/*
94 		 * return infinity
95 		 */
96 		Dbl_copytoptr(leftp1,leftp2,dstptr);
97 		return(NOEXCEPTION);
98 		}
99 	    }
100 	else
101 	    {
102 	    /*
103 	     * is NaN; signaling or quiet?
104 	     */
105 	    if (Dbl_isone_signaling(leftp1))
106 		{
107 		/* trap if INVALIDTRAP enabled */
108 		if (Is_invalidtrap_enabled()) return(INVALIDEXCEPTION);
109 		/* make NaN quiet */
110 		Set_invalidflag();
111 		Dbl_set_quiet(leftp1);
112 	    }
113 	    /*
114 	     * is second operand a signaling NaN?
115 	     */
116 	    else if (Dbl_is_signalingnan(rightp1))
117 		{
118 		/* trap if INVALIDTRAP enabled */
119 		if (Is_invalidtrap_enabled()) return(INVALIDEXCEPTION);
120 		/* make NaN quiet */
121 		Set_invalidflag();
122 		Dbl_set_quiet(rightp1);
123 		Dbl_copytoptr(rightp1,rightp2,dstptr);
124 		return(NOEXCEPTION);
125 		}
126 	    /*
127 	     * return quiet NaN
128 	     */
129 	    Dbl_copytoptr(leftp1,leftp2,dstptr);
130 	    return(NOEXCEPTION);
131 	    }
132 	} /* End left NaN or Infinity processing */
133     /*
134      * check second operand for NaN's or infinity
135      */
136     if (Dbl_isinfinity_exponent(rightp1))
137 	{
138 	if (Dbl_iszero_mantissa(rightp1,rightp2))
139 	    {
140 	    /* return infinity */
141 	    Dbl_copytoptr(rightp1,rightp2,dstptr);
142 	    return(NOEXCEPTION);
143 	    }
144 	/*
145 	 * is NaN; signaling or quiet?
146 	 */
147 	if (Dbl_isone_signaling(rightp1))
148 	    {
149 	    /* trap if INVALIDTRAP enabled */
150 	    if (Is_invalidtrap_enabled()) return(INVALIDEXCEPTION);
151 	    /* make NaN quiet */
152 	    Set_invalidflag();
153 	    Dbl_set_quiet(rightp1);
154 	    }
155 	/*
156 	 * return quiet NaN
157 	 */
158 	Dbl_copytoptr(rightp1,rightp2,dstptr);
159 	return(NOEXCEPTION);
160 	} /* End right NaN or Infinity processing */
161 
162     /* Invariant: Must be dealing with finite numbers */
163 
164     /* Compare operands by removing the sign */
165     Dbl_copytoint_exponentmantissap1(leftp1,signless_upper_left);
166     Dbl_copytoint_exponentmantissap1(rightp1,signless_upper_right);
167 
168     /* sign difference selects add or sub operation. */
169     if(Dbl_ismagnitudeless(leftp2,rightp2,signless_upper_left,signless_upper_right))
170 	{
171 	/* Set the left operand to the larger one by XOR swap	*
172 	 *  First finish the first word using "save"		*/
173 	Dbl_xorfromintp1(save,rightp1,/*to*/rightp1);
174 	Dbl_xorfromintp1(save,leftp1,/*to*/leftp1);
175 	Dbl_swap_lower(leftp2,rightp2);
176 	result_exponent = Dbl_exponent(leftp1);
177 	}
178     /* Invariant:  left is not smaller than right. */
179 
180     if((right_exponent = Dbl_exponent(rightp1)) == 0)
181 	{
182 	/* Denormalized operands.  First look for zeroes */
183 	if(Dbl_iszero_mantissa(rightp1,rightp2))
184 	    {
185 	    /* right is zero */
186 	    if(Dbl_iszero_exponentmantissa(leftp1,leftp2))
187 		{
188 		/* Both operands are zeros */
189 		if(Is_rounding_mode(ROUNDMINUS))
190 		    {
191 		    Dbl_or_signs(leftp1,/*with*/rightp1);
192 		    }
193 		else
194 		    {
195 		    Dbl_and_signs(leftp1,/*with*/rightp1);
196 		    }
197 		}
198 	    else
199 		{
200 		/* Left is not a zero and must be the result.  Trapped
201 		 * underflows are signaled if left is denormalized.  Result
202 		 * is always exact. */
203 		if( (result_exponent == 0) && Is_underflowtrap_enabled() )
204 		    {
205 		    /* need to normalize results mantissa */
206 		    sign_save = Dbl_signextendedsign(leftp1);
207 		    Dbl_leftshiftby1(leftp1,leftp2);
208 		    Dbl_normalize(leftp1,leftp2,result_exponent);
209 		    Dbl_set_sign(leftp1,/*using*/sign_save);
210 		    Dbl_setwrapped_exponent(leftp1,result_exponent,unfl);
211 		    Dbl_copytoptr(leftp1,leftp2,dstptr);
212 		    /* inexact = false */
213 		    return(UNDERFLOWEXCEPTION);
214 		    }
215 		}
216 	    Dbl_copytoptr(leftp1,leftp2,dstptr);
217 	    return(NOEXCEPTION);
218 	    }
219 
220 	/* Neither are zeroes */
221 	Dbl_clear_sign(rightp1);	/* Exponent is already cleared */
222 	if(result_exponent == 0 )
223 	    {
224 	    /* Both operands are denormalized.  The result must be exact
225 	     * and is simply calculated.  A sum could become normalized and a
226 	     * difference could cancel to a true zero. */
227 	    if( (/*signed*/int) save < 0 )
228 		{
229 		Dbl_subtract(leftp1,leftp2,/*minus*/rightp1,rightp2,
230 		/*into*/resultp1,resultp2);
231 		if(Dbl_iszero_mantissa(resultp1,resultp2))
232 		    {
233 		    if(Is_rounding_mode(ROUNDMINUS))
234 			{
235 			Dbl_setone_sign(resultp1);
236 			}
237 		    else
238 			{
239 			Dbl_setzero_sign(resultp1);
240 			}
241 		    Dbl_copytoptr(resultp1,resultp2,dstptr);
242 		    return(NOEXCEPTION);
243 		    }
244 		}
245 	    else
246 		{
247 		Dbl_addition(leftp1,leftp2,rightp1,rightp2,
248 		/*into*/resultp1,resultp2);
249 		if(Dbl_isone_hidden(resultp1))
250 		    {
251 		    Dbl_copytoptr(resultp1,resultp2,dstptr);
252 		    return(NOEXCEPTION);
253 		    }
254 		}
255 	    if(Is_underflowtrap_enabled())
256 		{
257 		/* need to normalize result */
258 		sign_save = Dbl_signextendedsign(resultp1);
259 		Dbl_leftshiftby1(resultp1,resultp2);
260 		Dbl_normalize(resultp1,resultp2,result_exponent);
261 		Dbl_set_sign(resultp1,/*using*/sign_save);
262 		Dbl_setwrapped_exponent(resultp1,result_exponent,unfl);
263 		Dbl_copytoptr(resultp1,resultp2,dstptr);
264 		/* inexact = false */
265 		return(UNDERFLOWEXCEPTION);
266 		}
267 	    Dbl_copytoptr(resultp1,resultp2,dstptr);
268 	    return(NOEXCEPTION);
269 	    }
270 	right_exponent = 1;	/* Set exponent to reflect different bias
271 				 * with denormalized numbers. */
272 	}
273     else
274 	{
275 	Dbl_clear_signexponent_set_hidden(rightp1);
276 	}
277     Dbl_clear_exponent_set_hidden(leftp1);
278     diff_exponent = result_exponent - right_exponent;
279 
280     /*
281      * Special case alignment of operands that would force alignment
282      * beyond the extent of the extension.  A further optimization
283      * could special case this but only reduces the path length for this
284      * infrequent case.
285      */
286     if(diff_exponent > DBL_THRESHOLD)
287 	{
288 	diff_exponent = DBL_THRESHOLD;
289 	}
290 
291     /* Align right operand by shifting to right */
292     Dbl_right_align(/*operand*/rightp1,rightp2,/*shifted by*/diff_exponent,
293     /*and lower to*/extent);
294 
295     /* Treat sum and difference of the operands separately. */
296     if( (/*signed*/int) save < 0 )
297 	{
298 	/*
299 	 * Difference of the two operands.  Their can be no overflow.  A
300 	 * borrow can occur out of the hidden bit and force a post
301 	 * normalization phase.
302 	 */
303 	Dbl_subtract_withextension(leftp1,leftp2,/*minus*/rightp1,rightp2,
304 	/*with*/extent,/*into*/resultp1,resultp2);
305 	if(Dbl_iszero_hidden(resultp1))
306 	    {
307 	    /* Handle normalization */
308 	    /* A straight forward algorithm would now shift the result
309 	     * and extension left until the hidden bit becomes one.  Not
310 	     * all of the extension bits need participate in the shift.
311 	     * Only the two most significant bits (round and guard) are
312 	     * needed.  If only a single shift is needed then the guard
313 	     * bit becomes a significant low order bit and the extension
314 	     * must participate in the rounding.  If more than a single
315 	     * shift is needed, then all bits to the right of the guard
316 	     * bit are zeros, and the guard bit may or may not be zero. */
317 	    sign_save = Dbl_signextendedsign(resultp1);
318 	    Dbl_leftshiftby1_withextent(resultp1,resultp2,extent,resultp1,resultp2);
319 
320 	    /* Need to check for a zero result.  The sign and exponent
321 	     * fields have already been zeroed.  The more efficient test
322 	     * of the full object can be used.
323 	     */
324 	    if(Dbl_iszero(resultp1,resultp2))
325 		/* Must have been "x-x" or "x+(-x)". */
326 		{
327 		if(Is_rounding_mode(ROUNDMINUS)) Dbl_setone_sign(resultp1);
328 		Dbl_copytoptr(resultp1,resultp2,dstptr);
329 		return(NOEXCEPTION);
330 		}
331 	    result_exponent--;
332 	    /* Look to see if normalization is finished. */
333 	    if(Dbl_isone_hidden(resultp1))
334 		{
335 		if(result_exponent==0)
336 		    {
337 		    /* Denormalized, exponent should be zero.  Left operand *
338 		     * was normalized, so extent (guard, round) was zero    */
339 		    goto underflow;
340 		    }
341 		else
342 		    {
343 		    /* No further normalization is needed. */
344 		    Dbl_set_sign(resultp1,/*using*/sign_save);
345 		    Ext_leftshiftby1(extent);
346 		    goto round;
347 		    }
348 		}
349 
350 	    /* Check for denormalized, exponent should be zero.  Left    *
351 	     * operand was normalized, so extent (guard, round) was zero */
352 	    if(!(underflowtrap = Is_underflowtrap_enabled()) &&
353 	       result_exponent==0) goto underflow;
354 
355 	    /* Shift extension to complete one bit of normalization and
356 	     * update exponent. */
357 	    Ext_leftshiftby1(extent);
358 
359 	    /* Discover first one bit to determine shift amount.  Use a
360 	     * modified binary search.  We have already shifted the result
361 	     * one position right and still not found a one so the remainder
362 	     * of the extension must be zero and simplifies rounding. */
363 	    /* Scan bytes */
364 	    while(Dbl_iszero_hiddenhigh7mantissa(resultp1))
365 		{
366 		Dbl_leftshiftby8(resultp1,resultp2);
367 		if((result_exponent -= 8) <= 0  && !underflowtrap)
368 		    goto underflow;
369 		}
370 	    /* Now narrow it down to the nibble */
371 	    if(Dbl_iszero_hiddenhigh3mantissa(resultp1))
372 		{
373 		/* The lower nibble contains the normalizing one */
374 		Dbl_leftshiftby4(resultp1,resultp2);
375 		if((result_exponent -= 4) <= 0 && !underflowtrap)
376 		    goto underflow;
377 		}
378 	    /* Select case were first bit is set (already normalized)
379 	     * otherwise select the proper shift. */
380 	    if((jumpsize = Dbl_hiddenhigh3mantissa(resultp1)) > 7)
381 		{
382 		/* Already normalized */
383 		if(result_exponent <= 0) goto underflow;
384 		Dbl_set_sign(resultp1,/*using*/sign_save);
385 		Dbl_set_exponent(resultp1,/*using*/result_exponent);
386 		Dbl_copytoptr(resultp1,resultp2,dstptr);
387 		return(NOEXCEPTION);
388 		}
389 	    Dbl_sethigh4bits(resultp1,/*using*/sign_save);
390 	    switch(jumpsize)
391 		{
392 		case 1:
393 		    {
394 		    Dbl_leftshiftby3(resultp1,resultp2);
395 		    result_exponent -= 3;
396 		    break;
397 		    }
398 		case 2:
399 		case 3:
400 		    {
401 		    Dbl_leftshiftby2(resultp1,resultp2);
402 		    result_exponent -= 2;
403 		    break;
404 		    }
405 		case 4:
406 		case 5:
407 		case 6:
408 		case 7:
409 		    {
410 		    Dbl_leftshiftby1(resultp1,resultp2);
411 		    result_exponent -= 1;
412 		    break;
413 		    }
414 		}
415 	    if(result_exponent > 0)
416 		{
417 		Dbl_set_exponent(resultp1,/*using*/result_exponent);
418 		Dbl_copytoptr(resultp1,resultp2,dstptr);
419 		return(NOEXCEPTION);	/* Sign bit is already set */
420 		}
421 	    /* Fixup potential underflows */
422 	  underflow:
423 	    if(Is_underflowtrap_enabled())
424 		{
425 		Dbl_set_sign(resultp1,sign_save);
426 		Dbl_setwrapped_exponent(resultp1,result_exponent,unfl);
427 		Dbl_copytoptr(resultp1,resultp2,dstptr);
428 		/* inexact = false */
429 		return(UNDERFLOWEXCEPTION);
430 		}
431 	    /*
432 	     * Since we cannot get an inexact denormalized result,
433 	     * we can now return.
434 	     */
435 	    Dbl_fix_overshift(resultp1,resultp2,(1-result_exponent),extent);
436 	    Dbl_clear_signexponent(resultp1);
437 	    Dbl_set_sign(resultp1,sign_save);
438 	    Dbl_copytoptr(resultp1,resultp2,dstptr);
439 	    return(NOEXCEPTION);
440 	    } /* end if(hidden...)... */
441 	/* Fall through and round */
442 	} /* end if(save < 0)... */
443     else
444 	{
445 	/* Add magnitudes */
446 	Dbl_addition(leftp1,leftp2,rightp1,rightp2,/*to*/resultp1,resultp2);
447 	if(Dbl_isone_hiddenoverflow(resultp1))
448 	    {
449 	    /* Prenormalization required. */
450 	    Dbl_rightshiftby1_withextent(resultp2,extent,extent);
451 	    Dbl_arithrightshiftby1(resultp1,resultp2);
452 	    result_exponent++;
453 	    } /* end if hiddenoverflow... */
454 	} /* end else ...add magnitudes... */
455 
456     /* Round the result.  If the extension is all zeros,then the result is
457      * exact.  Otherwise round in the correct direction.  No underflow is
458      * possible. If a postnormalization is necessary, then the mantissa is
459      * all zeros so no shift is needed. */
460   round:
461     if(Ext_isnotzero(extent))
462 	{
463 	inexact = true;
464 	switch(Rounding_mode())
465 	    {
466 	    case ROUNDNEAREST: /* The default. */
467 	    if(Ext_isone_sign(extent))
468 		{
469 		/* at least 1/2 ulp */
470 		if(Ext_isnotzero_lower(extent)  ||
471 		  Dbl_isone_lowmantissap2(resultp2))
472 		    {
473 		    /* either exactly half way and odd or more than 1/2ulp */
474 		    Dbl_increment(resultp1,resultp2);
475 		    }
476 		}
477 	    break;
478 
479 	    case ROUNDPLUS:
480 	    if(Dbl_iszero_sign(resultp1))
481 		{
482 		/* Round up positive results */
483 		Dbl_increment(resultp1,resultp2);
484 		}
485 	    break;
486 
487 	    case ROUNDMINUS:
488 	    if(Dbl_isone_sign(resultp1))
489 		{
490 		/* Round down negative results */
491 		Dbl_increment(resultp1,resultp2);
492 		}
493 
494 	    case ROUNDZERO:;
495 	    /* truncate is simple */
496 	    } /* end switch... */
497 	if(Dbl_isone_hiddenoverflow(resultp1)) result_exponent++;
498 	}
499     if(result_exponent == DBL_INFINITY_EXPONENT)
500 	{
501 	/* Overflow */
502 	if(Is_overflowtrap_enabled())
503 	    {
504 	    Dbl_setwrapped_exponent(resultp1,result_exponent,ovfl);
505 	    Dbl_copytoptr(resultp1,resultp2,dstptr);
506 	    if (inexact) {
507 		if (Is_inexacttrap_enabled())
508 			return(OVERFLOWEXCEPTION | INEXACTEXCEPTION);
509 		else
510 			Set_inexactflag();
511 	    }
512 	    return(OVERFLOWEXCEPTION);
513 	    }
514 	else
515 	    {
516 	    inexact = true;
517 	    Set_overflowflag();
518 	    Dbl_setoverflow(resultp1,resultp2);
519 	    }
520 	}
521     else Dbl_set_exponent(resultp1,result_exponent);
522     Dbl_copytoptr(resultp1,resultp2,dstptr);
523     if(inexact) {
524 	if(Is_inexacttrap_enabled())
525 	    return(INEXACTEXCEPTION);
526 	else
527 	    Set_inexactflag();
528     }
529     return(NOEXCEPTION);
530     }
531